Cortex: Content Analysis Support Software, a Resource for Qualitative Research

# Cortex: Content Analysis Support Software, a Resource for Qualitative Research
Source: [https://arxiv.org/html/2609.11970](https://arxiv.org/html/2609.11970)
Ana Julia da Silva SoaresAffiliation:Instituto Federal de Educação, Ciência e Tecnologia do Rio Grande do Sul \(IFRS\), Campus Canoas, BrazilRafael Coimbra PintoAffiliation:Instituto Federal de Educação, Ciência e Tecnologia do Rio Grande do Sul \(IFRS\), Campus Canoas, Brazil

2026

###### Abstract

Qualitative research is widely used in the human and social sciences, characterized by a deep understanding of phenomena through the interpretation of meanings and contexts\. Among qualitative data analysis methods, content analysis stands out as a consolidated technique, which allows for the systematic description and interpretation of textual contents\. However, as data volume increases, the time required for organization, reading, and categorization becomes a significant challenge, potentially delaying research development\. Therefore, this work aimed to develop a web application to support content analysis, based on Bardin’s methodology, targeted at academic researchers\. The methodology adopted a mixed approach, combining bibliographic research on content analysis with semi\-structured interviews with four experienced researchers, aiming to identify real needs and requirements\. Based on these inputs, the Cortex software was developed to assist the researcher in the pre\-analysis and material exploration stages, generating suggestions for indices, indicators, and categories with full traceability to original documents\. The results demonstrate that Cortex is capable of guiding the researcher through the entire methodological workflow, from corpus configuration to results exportation, acting as a methodological collaborator without replacing the researcher’s interpretative autonomy\.

Keywords:Content Analysis; Qualitative Research; Artificial Intelligence; Natural Language Processing; RAG\.

## 1Introduction

Qualitative research is a methodological approach widely used in human and social sciences, characterized by its focus on the deep understanding of social, cultural, and subjective phenomena\[[2](https://arxiv.org/html/2609.11970#bib.bib2)\]\. According to Godoy\[[7](https://arxiv.org/html/2609.11970#bib.bib7)\], this type of research is not based on the quantification of events or the application of statistical instruments, but starts from broad questions that are delineated throughout the study\. The author highlights that such research seeks to describe people, contexts, and processes based on the direct interaction between the researcher and the object of study, prizing the understanding of phenomena from the perspective of the subjects involved\[[7](https://arxiv.org/html/2609.11970#bib.bib7)\]\. Unlike quantitative approaches designed to measure variables, qualitative studies rely on the interpretation of meanings, contexts, and experiences, employing methods such as interviews, observations, and discourse analysis\[[11](https://arxiv.org/html/2609.11970#bib.bib11)\]\.

Among the methods used to analyze textual data obtained in this domain, Content Analysis stands out as a consolidated and effective technique originating in the late 20th century\. According to Moraes\[[12](https://arxiv.org/html/2609.11970#bib.bib12)\], it is a methodology that allows describing and interpreting the contents of different types of documents and texts, enabling a deep understanding of the core messages\. When applied with rigor, this approach transforms raw data into structured knowledge, being widely recognized for its theoretical and methodological consistency\[[16](https://arxiv.org/html/2609.11970#bib.bib16)\]\.

However, as the volume of data increases — especially in contemporary studies involving multiple dense interviews or large\-scale open\-ended responses — the time and cognitive effort required for organizing, reading, and analyzing content become significant challenges\[[16](https://arxiv.org/html/2609.11970#bib.bib16)\]\. Coder fatigue, tight deadlines, and the inherent complexity of manual coding rules can compromise analysis precision and introduce subjective biases\[[15](https://arxiv.org/html/2609.11970#bib.bib15)\]\. Bardin\[[1](https://arxiv.org/html/2609.11970#bib.bib1)\]emphasizes that maintaining an unbiased and systematic analytical stance demands a continuous psychological effort from the analyst, stating that:

> "structural deciphering centered on each interview \[requires\] an effort but that does not exclude intuition, to the extent that, in each new interview, it is necessary to make an abstraction of oneself and previous interviews \[…\], to make a clean slate, a priori, of personal opinions or contamination from previous decipherings"\[[1](https://arxiv.org/html/2609.11970#bib.bib1), p\. 96\]\.

Considering that qualitative analysis is only one stage of the research process, the time spent in this phase can delay the general development of the study\. Therefore, tools that contribute to speeding up this stage, without compromising the quality of the analysis, can represent an important support to scientific activity\.

To address these limitations, this work presents Cortex, a web application designed to support academic researchers throughout the explicit chronological phases of Content Analysis, according to the methodology proposed by Bardin\[[1](https://arxiv.org/html/2609.11970#bib.bib1)\]\. The tool allows users to import their textual research corpus and theoretical frameworks to automatically generate suggestions for indices, indicators, and thematic categories via a Large Language Model \(LLM\), while maintaining strict verification and traceability to the original source text\.

## 2Methodology and System Design

This work adopts a mixed approach, combining quantitative and qualitative methods, as suggested by Creswell and Plano Clark\[[4](https://arxiv.org/html/2609.11970#bib.bib4)\], focusing on the development of a web application to support researchers in the qualitative content analysis process\.

Initially, a bibliographic search was carried out on content analysis methodology and its applications in qualitative research to theoretically base the tool’s development\. In parallel, four semi\-structured interviews were conducted with researchers who have experience with content analysis methodology or qualitative methods in general\.

The group of participants consisted of professors and researchers linked to higher education institutions, all with postgraduate degrees \(master’s and doctorate\)\. The multidisciplinarity of the group, covering distinct areas such as Computer Science, International Relations, and Education, was instrumental in capturing diverse cross\-domain challenges\.

The objective of this stage was to identify needs, difficulties, and expectations regarding the use of digital tools in the process of categorization, visualization, and analysis of textual data\. The interviews were transcribed and analyzed qualitatively, allowing the identification of requirements aligned with the real demands of future application users\.

Based on the requirements and inputs obtained in this initial phase, the system was modeled to graphically represent the main functionalities and structural entities\. The database modeling and the domain classes of Cortex were designed to reflect the Content Analysis methodology proposed by Bardin\[[1](https://arxiv.org/html/2609.11970#bib.bib1)\]\. Each methodological stage is represented by specific entities that store the artifacts generated in each phase\. Consequently, the core structure includes strictly defined entities for the Analysis, the Document, the Index, the Indicator, the Category, and the Register Unit\.

Once modeling was completed and requirements defined, the software development phase began\. To operationalize this theoretical framework, the software architecture was divided into four cohesive functional modules: Authentication, Configuration, Pre\-Analysis, and Exploration\.

The application was built using a decoupled client\-server model, utilizing C\# with \.NET in the backend to provide a robust, high\-performance API\. The frontend was developed using React with TypeScript, ensuring a responsive user interface capable of managing dynamic data visualizations\. The structural organization and data flow between these components are illustrated in Figure[1](https://arxiv.org/html/2609.11970#S2.F1)\.

![Refer to caption](https://arxiv.org/html/2609.11970v1/architecture.png)Figure 1:High\-level architecture of the Cortex system, illustrating the interaction between the client interface, the backend API, hybrid persistence layers, and the LLM RAG mechanism\.To elucidate the dynamic behavior of Cortex, the data flow adopts a hybrid storage strategy during the configuration stage, which corresponds to the constitution of the corpus\. Physical files, such as PDFs or text documents, are sent directly to Google Cloud Storage to guarantee scalability, while the local PostgreSQL database stores only the metadata and the logical references, or URIs, returned by the cloud\.

For the intelligent processing phases, the system integrates the Gemini model through its API on the Vertex AI platform, applying the Retrieval\-Augmented Generation \(RAG\) technique\. In both the Pre\-Analysis and Exploration stages, the backend queries the database for the document URIs, elaborates a methodological prompt, and defines a strict JSON Schema\. These inputs are sent to Vertex AI, which reads the original documents and generates a structured response using Natural Language Processing \(NLP\) techniques\[[17](https://arxiv.org/html/2609.11970#bib.bib17)\]\. Finally, the LLM’s response is validated by the schema, automatically deserialized into C\# domain classes, and persisted in the relational database, allowing the frontend to render the final results in comprehensive tables and graphs\.

## 3Theoretical Background

### 3\.1Qualitative Research and Content Analysis

Qualitative research is a methodology that is not concerned with numerical representativeness, but with deepening the understanding of a social group, an organization, or phenomena that cannot be quantified, focusing on explaining the dynamics of social relations\[[6](https://arxiv.org/html/2609.11970#bib.bib6)\]\. According to Godoy\[[7](https://arxiv.org/html/2609.11970#bib.bib7)\], qualitative research does not seek to enumerate or measure events, nor does it employ statistical instruments in data analysis, but rather starts from broad questions that are defined throughout the study\.

Content analysis aligns directly with qualitative research, defined as a methodology of this nature, especially in the field of Social Sciences\[[3](https://arxiv.org/html/2609.11970#bib.bib3)\], but which can also be used to analyze quantitative data, according to Bardin\[[1](https://arxiv.org/html/2609.11970#bib.bib1)\]\. This approach goes beyond the simple description of qualitative data, allowing new interpretations that deepen the understanding of underlying meanings present in collected information\[[12](https://arxiv.org/html/2609.11970#bib.bib12)\]\.

Because of this deeply interpretative nature, Minayo\[[11](https://arxiv.org/html/2609.11970#bib.bib11)\]notes that content analysis inevitably involves the researcher’s personal lens\. She argues that a completely neutral, objective, and exhaustive reading is unattainable, as the values, language, and cultural context of both the researcher and the object of study actively influence data interpretation\.

To mitigate this inherent subjectivity and ensure scientific rigor, the methodology must follow a highly structured workflow\. In this regard, Bardin\[[1](https://arxiv.org/html/2609.11970#bib.bib1)\]organizes the phases of Content Analysis around three explicit chronological poles: pre\-analysis; exploration of the material; and treatment of results, inference, and interpretation\.

#### 3\.1\.1Pre\-analysis

This is the initial phase of organization, intuition, and systematization of ideas\. In this phase \(also called corpus constitution\), the choice of documents to be analyzed is made\. According to Bardin\[[1](https://arxiv.org/html/2609.11970#bib.bib1)\], the first sub\-stage is "floating reading", where the researcher comes into contact with the material\.

Still in this stage, preparation for future data interpretation occurs through the elaboration of indicators\. Bardin describes the corpus as a manifestation carrying ’vestiges’ \(indices\) that the analysis must ’make speak’\. The researcher’s task is to identify these indices on the textual surface and organize them systematically into indicators\. While indices are concrete signs in the text, indicators represent the measurement rule or method adopted to quantify \(or qualify\) these signs\[[3](https://arxiv.org/html/2609.11970#bib.bib3)\]\.

To illustrate this distinction, Bardin\[[1](https://arxiv.org/html/2609.11970#bib.bib1)\]uses a classic example of a therapeutic interview\. The author proposes a hypothesis where emotion and anxiety \(the concept to be inferred\) would manifest through speech disturbances\. In this scenario, the indices \(vestiges\) are the concrete signs in communication, such as interrupted sentences, stuttering, or incoherent sounds \(e\.g\., "uh", "um"\)\. The indicator \(measure\), in turn, is the counting rule defined by the researcher, such as the frequency of appearance of these indices\. It is this transformation of an abstract phenomenon into a concrete indicator that allows the analysis to be objective and systematic\. Table[1](https://arxiv.org/html/2609.11970#S3.T1)summarizes this relationship\.

Table 1:Example of the relationship between hypothesis, indices, and indicators in content analysisConcept to be Inferred \(Inferred Variable\)HypothesisIndices \(Vestiges in Text\)Indicator \(Systematic Measure\)Anxiety/EmotionAnxiety manifests through speech disturbancesInterrupted sentences, word repetition, stuttering, incoherent sounds \(e\.g\., "uh", "um"\)The frequency of appearance of these indices
Source: Elaborated by the author, based on Bardin \(2011\)\.

#### 3\.1\.2Exploration of the Material

The material exploration stage consists of the systematic application of the decisions made during pre\-analysis\. According to Bardin\[[1](https://arxiv.org/html/2609.11970#bib.bib1)\], this is the phase where the collected material is effectively transformed into analyzable data through coding operations\. With the objectives, corpus, and indicators already defined, the researcher applies the analysis techniques to the text\.

The central operation of this phase is material coding\. Before classifying excerpts from the corpus, the text must be broken down into smaller, manageable segments\. The most common technique, Categorical Analysis \(or Thematic Analysis\), begins by dividing the text into two distinct units: the recording unit and the context unit\[[1](https://arxiv.org/html/2609.11970#bib.bib1)\]\.

The recording unit is the core unit of meaning \(e\.g\., a specific theme or word\) that the researcher will register and count\. Conversely, the context unit is the larger text segment surrounding the recording unit, providing the necessary background to fully comprehend its exact meaning\. Context is crucial in qualitative analysis, as the true significance of isolated elements heavily depends on the broader discourse\[[1](https://arxiv.org/html/2609.11970#bib.bib1)\]\.

After segmenting the text, the analyst groups these units into categories\. Categories act as conceptual "drawers" or significant rubrics that allow the classification of the message’s constitutive meaning elements\. Once categorization is established, the researcher’s task is to count the frequency of presence \(or absence\) of these meaning items within each category\. In Categorical Analysis, the frequency with which certain content features appear serves as the primary analytical information\.

Given the highly systematic nature of this phase—which involves meticulous coding, categorization, and frequency counting—Bardin\[[1](https://arxiv.org/html/2609.11970#bib.bib1)\]highlighted that the use of computers is a remarkably valuable resource\. The author emphasized their utility for complex analyses involving multiple variables, occurrence tracking, and elaborated statistical operations, significantly accelerating a process that would otherwise be exhaustingly manual\. This precise methodological gap is the primary target of the Cortex system’s automation capabilities\.

#### 3\.1\.3Treatment of Results, Inference, and Interpretation

In this phase, raw data is processed to highlight and interpret the obtained information\[[6](https://arxiv.org/html/2609.11970#bib.bib6)\]\. The results can be condensed and visually represented through diagrams, figures, and models\[[3](https://arxiv.org/html/2609.11970#bib.bib3)\]\.

Based on these results, inferences can be made, which represent the essential purpose of this methodology\. As Bardin explains, "the intention of content analysis is the inference of knowledge related to the conditions of production \(or, eventually, reception\), an inference that relies on indicators \(quantitative or otherwise\)"\[[1](https://arxiv.org/html/2609.11970#bib.bib1), p\. 44\]\.

Interpretation \(the assignment of meaning to the detected characteristics\) is the final stage, while inference serves as the intermediate procedure that allows an explicit and controlled transition from the description phase to the interpretation\[[1](https://arxiv.org/html/2609.11970#bib.bib1)\]\.

This link between objective description \(the counted data\) and interpretation \(the assigned meaning\) is bridged by the theoretical framework\. Therefore, inference is not random; it consists of analyzing the descriptive results in light of the hypotheses and theories that grounded the pre\-analysis\. It is this return to theory that allows the researcher to propose meanings for the findings, going beyond what is merely explicit in the text\[[1](https://arxiv.org/html/2609.11970#bib.bib1)\]\.

Finally, the validity of Content Analysis should not be judged against a single "true reading" of the text — since the meaning of a text is rarely entirely manifest or unique — but rather by its grounding in the researched materials, its congruence with the researcher’s theoretical framework, and its alignment with the research objective\[[3](https://arxiv.org/html/2609.11970#bib.bib3)\]\.

### 3\.2Natural Language Processing and Artificial Intelligence

Natural Language Processing \(NLP\) is a subfield of artificial intelligence focused on enabling computers to understand and process human languages\[[9](https://arxiv.org/html/2609.11970#bib.bib9)\]\. It studies fundamental technologies for expressing word, phrase, sentence, and document meanings\. The most significant advances in NLP in recent decades have been driven by AI, particularly Machine Learning \(ML\) and Deep Learning\.

### 3\.3Retrieval\-Augmented Generation \(RAG\)

While Content Analysis is a human inference methodological process, applying computational NLP techniques allows automating parts of this process\. Large Language Models \(LLMs\) have become prominent, but they have limitations such as hallucinations and knowledge cutoffs\[[10](https://arxiv.org/html/2609.11970#bib.bib10)\]\.

To solve these limitations, Retrieval\-Augmented Generation \(RAG\) architecture was developed\. This technique transforms the LLM from a closed knowledge agent to a system that reasons based on external and updated information sources\. It consists fundamentally of two phases: Retrieval and Generation\.

For the RAG system to work, three technical components are fundamental: Embeddings \(Vectorization\), Chunking \(Fragmentation\), and a Similarity Search mechanism\. The process begins with chunking, which divides extensive documents into smaller, semantically cohesive fragments\. Then, embeddings are used to create a numerical representation of this data in a high\-dimensional vector space\. Finally, similarity retrieval compares the user query vector with the chunk vectors to find the relevant context\.

#### 3\.3\.1Vertex AI

The implementation of a RAG system involves the orchestration of all the aforementioned components: document preprocessing, chunking, embedding generation, vector indexing, and the call to the LLM\[[10](https://arxiv.org/html/2609.11970#bib.bib10)\]\. Cloud platforms, such as Google Cloud’s Vertex AI, offer managed services that abstract this complexity\.

According to the official documentation\[[8](https://arxiv.org/html/2609.11970#bib.bib8)\], the Vertex AI platform allows users to provide their source documents \(such as PDFs and TXTs\)\. The tool then automatically manages the data ingestion pipeline, which includes parsing, chunking, and indexing \(embedding creation\) the data\. Once indexed, the corpus becomes available to be queried by the Gemini model, which utilizes the platform’s RAG mechanism to retrieve relevant excerpts and ground its responses\[[8](https://arxiv.org/html/2609.11970#bib.bib8)\]\. Therefore, utilizing this API enables the application of the RAG technique without the need to manually build and maintain each component of the pipeline\.

## 4Related Work

### 4\.1ATLAS\.ti

ATLAS\.ti is one of the most consolidated qualitative analysis software in academia\. Its structure is influenced by Grounded Theory\. The workflow relies on creating a "project" where the researcher imports documents and performs "coding" \(selecting text excerpts and assigning codes\)\. Its strength lies in managing and visualizing relationships between codes \(semantic networks\)\. Recently, it incorporated LLM\-powered features, but its core remains a sophisticated "workbench" for manual process\[[13](https://arxiv.org/html/2609.11970#bib.bib13)\]\.

### 4\.2NVivo

NVivo is the main competitor to ATLAS\.ti\. It allows coding text excerpts into "Nodes"\. Its query resources are a differentiator, allowing complex searches and coding matrices\. Like its competitor, it has introduced LLM modules, but requires the researcher to read, interpret, and code material manually before inferring\[[14](https://arxiv.org/html/2609.11970#bib.bib14)\]\.

### 4\.3Elicit

Elicit represents a paradigm shift\. It is not a traditional CAQDAS, but an "AI research assistant" native to the LLM era\. Its main focus is automating research tasks, especially literature review\. Using RAG, Elicit allows users to ask questions in natural language, and the tool locates scientific articles and extracts key information\. However, its design is not aimed at analyzing primary data \(like interviews\) under a specific qualitative methodology\[[5](https://arxiv.org/html/2609.11970#bib.bib5)\]\.

### 4\.4Comparative Analysis and Gap Identification

The analysis of the tools demonstrates clear differences\. On one hand, ATLAS\.ti and NVivo offer robust environments for manual methodological rigor, excelling at managing the complexity of coding, but they have a steep learning curve and do not automate the cognitive phases of the analysis\. On the other hand, Elicit uses advanced LLMs to automate synthesis but focuses on literature review and does not adapt to the prescriptive steps of methodologies such as Bardin’s\[[1](https://arxiv.org/html/2609.11970#bib.bib1)\]\.

None of the analyzed tools were designed to fill the central gap identified by this work: the application of Large Language Models \(integrated with the RAG architecture\) to actively assist the researcher within the specific phases of Content Analysis\.

While NVivo and ATLAS\.ti treat the researcher as the sole agent of inference, and Elicit acts as an automated literature synthesizer, the proposal of this work \(the Cortex software\) positions itself as a methodological collaborator\. The objective of Cortex is to use a Large Language Model via the Gemini API not to replace the researcher, but to accelerate the pre\-analysis and exploration stages by suggesting preliminary indices, indicators, and categories based on the provided corpus and theoretical framework\. Table[2](https://arxiv.org/html/2609.11970#S4.T2)below summarizes the main differences between the analyzed tools and the proposal of this work\.

Table 2:Comparative analysis of data analysis tools

## 5Results: The Cortex Software

The development of Cortex resulted in a robust web application that actively guides the qualitative researcher through the chronological phases of Content Analysis\. The system’s interface and underlying logic were designed to abstract the technical complexity of Large Language Models \(LLMs\), presenting a workflow entirely focused on methodological rigor\.

### 5\.1Authentication and Workspace Initialization

The researcher’s interaction begins in a secure, authenticated environment utilizing JSON Web Tokens \(JWT\) to ensure strict data isolation between users and projects\. Upon authentication, the main dashboard provides a tabular overview of all ongoing analyses\. Given the extensive time required for qualitative research, the system implements continuous persistence, allowing the user to pause their work at any methodological stage and resume later without data loss\. To initiate a new project, the researcher simply defines the analysis title, immediately advancing to the corpus constitution phase\.

### 5\.2Configuration Stage: Corpus Constitution

The configuration stage directly mirrors the initial corpus constitution described by Bardin\[[1](https://arxiv.org/html/2609.11970#bib.bib1)\]\. In this environment, the researcher establishes three fundamental elements\. First, they must define the central research question, which acts as the primary semantic anchor for all subsequent LLM processing\. Following this, the user uploads the textual documents to be analyzed, typically interview transcripts in PDF or TXT formats, which form the primary corpus\. Finally, the system allows the optional upload of reference documents, such as scientific articles or theoretical frameworks\. These reference files are not analyzed as primary data but are ingested by the RAG mechanism to contextualize and ground the LLM’s future suggestions\. Once these parameters are validated, the system strictly enforces linear progression, preventing backward alterations to maintain methodological integrity\.

### 5\.3Pre\-Analysis: LLM\-Driven Index Generation

Upon advancing, Cortex initiates the Pre\-Analysis stage by orchestrating the first interaction with the Gemini model\. The backend constructs a highly structured prompt incorporating the central research question, the corpus metadata, and the explicit criteria defined by Bardin\[[1](https://arxiv.org/html/2609.11970#bib.bib1)\]for index selection and indicator construction\. This prompt, along with the document URIs, is processed by Vertex AI using the RAG architecture\.

The LLM’s response is deserialized and presented to the researcher as interactive cards, as illustrated in Figure[2](https://arxiv.org/html/2609.11970#S5.F2)\. Each card displays a suggested index, its theoretical description, the associated indicator, and a critical list of exact textual references extracted directly from the uploaded documents, complete with page numbers\. To enhance traceability, an integrated document viewer positioned alongside the cards allows the researcher to simultaneously read the full original context of any generated reference in real\-time, validating the LLM’s inferences against the raw data\.

![Refer to caption](https://arxiv.org/html/2609.11970v1/preAnalysis.png)Figure 2:The Pre\-Analysis interface displaying LLM\-generated indices alongside the integrated document viewer\. Key visual elements include the suggested index name and description, the systematic counting rule, and the exact textual extractions linking the LLM’s conceptual suggestion directly to the source document and page number\.
### 5\.4Human\-in\-the\-Loop: Concept Appropriation

A critical design decision in Cortex is the enforcement of researcher autonomy over the LLM’s outputs\. During the Pre\-Analysis stage, the researcher exercises full methodological appropriation by editing the theoretical names and descriptions of the generated indices, removing irrelevant suggestions, or creating entirely new indices manually\. However, the system deliberately prevents the user from editing the raw textual references and page numbers extracted by the LLM\. This constraint guarantees that while the theoretical interpretation remains entirely under human control, the documentary foundation remains immutable, ensuring absolute analytical traceability\.

### 5\.5Material Exploration and Automated Categorization

The Material Exploration stage represents the most technically complex phase of the system, where the thematic categorization of the corpus occurs\. The system retrieves all the indices and indicators previously validated by the researcher and constructs a second structured prompt\. This new instruction commands the LLM to identify specific register units within the text and group them into cohesive categories based strictly on the validated indices and frequency counting rules\.

The results of this exploration are rendered in three complementary visualizations\. First, a graphical bar chart displays the frequency distribution of the categories, allowing for a rapid visual assessment of thematic representativeness\. Second, a tabular categorization matrix, as presented in Figure[3](https://arxiv.org/html/2609.11970#S5.F3), details the co\-occurrence frequencies between the identified categories and the applied indices\. Finally, an expandable card layout presents each category’s semantic definition, total frequency, and the exhaustive list of related register units\. Each register unit displays the exact textual excerpt, its source document, and a justification for its categorization\.

![Refer to caption](https://arxiv.org/html/2609.11970v1/tableCategories.png)Figure 3:The Material Exploration interface displaying the categorization matrix\. The table illustrates the final methodological output, cross\-referencing the generated Categories, their total frequencies, and the specific distribution of Found Indices within each thematic group\.Notably, unlike the Pre\-Analysis stage, these final categories are non\-editable\. This architectural restriction ensures methodological validity, as the categories must remain the direct, systematic mathematical result of applying the previously validated indices over the corpus\.

### 5\.6Results Export and Methodological Traceability

To conclude the analytical workflow, Cortex generates a comprehensive PDF report containing the entire analytical trail\. This final artifact encompasses the project’s metadata, the central research question, the frequency distribution graphs, and the complete categorization matrix\. Furthermore, it extensively details every category, its associated indices, the employed indicators, and an exhaustive list of all register units with their exact textual origins and justifications\. This exported report serves as a definitive, traceable record, enabling readers and peer reviewers to follow the exact methodological path from the raw textual corpus to the final conceptual inferences\.

## 6Final Considerations

This work developed a web application to support content analysis, grounded in Bardin’s methodology111The source code and repository for the Cortex project are publicly available at[https://github\.com/AnaJuliaSSdev/Cortex](https://github.com/AnaJuliaSSdev/Cortex)\.\. Cortex acts as a methodological assistant, automating laborious steps \(index identification, counting\) while preserving human control over interpretative decisions\. The development process followed specific objectives, from theoretical research to implementation using modern stack \(\.NET, React, Vertex AI\)\. Results show the system successfully guides researchers through the workflow\. Limitations include export restricted to PDF in the MVP\. Future work includes multi\-format export, audio transcription integration, and advanced visualizations \(word clouds\)\.

## References

- \[1\]Bardin, L\. \(2011\)\.Análise de conteúdo\. São Paulo: Edições 70\.
- \[2\]Breuer, F\., & Reichertz, J\. \(2002\)\. Standards of social research\.Forum: Qualitative Social Research, 2\(3\)\.
- \[3\]Cardoso, M\. R\. G\., Oliveira, G\. S\., & Ghelli, K\. G\. M\. \(2021\)\. Análise de conteúdo: uma metodologia de pesquisa qualitativa\.Cadernos da FUCAMP, 20\(43\), 92\-106\.
- \[4\]Creswell, J\. W\., & Plano Clark, V\. L\. \(2011\)\.Designing and Conducting Mixed Methods Research\. 2nd ed\. Thousand Oaks: SAGE\.
- \[5\]Elicit\. \(2024\)\.Elicit: The AI Research Assistant\. Available at: https://elicit\.com/\.
- \[6\]Gerhardt, T\. E\., & Silveira, D\. T\. \(orgs\.\)\. \(2009\)\.Métodos de pesquisa\. Porto Alegre: Editora da UFRGS\.
- \[7\]Godoy, A\. S\. \(1995\)\. Introdução à pesquisa qualitativa e suas possibilidades\.RAE \- Revista de Administração de Empresas, 35\(2\), 57\-63\.
- \[8\]Google\. \(2024\)\.Retrieval\-Augmented Generation \(RAG\) Overview\. Google Cloud Vertex AI\.
- \[9\]Howard, J\. \(2019\)\. Artificial intelligence: Implications for the future of work\.American Journal of Industrial Medicine, 62\(11\), 1\-10\.
- \[10\]James, A\., Trovati, M\., & Bolton, S\. \(2025\)\. Retrieval\-Augmented Generation to Generate Knowledge Assets\.Applied Sciences, 15, 6247\.
- \[11\]Minayo, M\. C\. S\. \(2001\)\.Pesquisa social: teoria, método e criatividade\. 18th ed\. Petrópolis: Vozes\.
- \[12\]Moraes, R\. \(1999\)\. Análise de conteúdo\.Revista Educação, 22\(37\), 7\-32\.
- \[13\]Muhr, T\. \(1991\)\. ATLAS\.ti: a prototype for the support of text interpretation\.Qualitative Sociology, 14\(4\), 349\-371\.
- \[14\]QSR International\. \(2014\)\.NVivo 10: Getting Started Guide\.
- \[15\]Sampaio, R\. C\., & Lycarião, D\. \(2021\)\.Análise de conteúdo categorial: manual de aplicação\. Brasília: Enap\.
- \[16\]Sousa, J\. R\., & Santos, S\. C\. M\. \(2020\)\. Análise de conteúdo em pesquisa qualitativa: modo de pensar e de fazer\.Pesquisa e Debate em Educação, 10\(2\), 1396\-1416\.
- \[17\]Zhou, M\. et al\. \(2020\)\. Progress in neural NLP: Modeling, learning, and reasoning\.Engineering, 6\(3\), 275\-290\.

## Appendix APrompts Used in the System

The following prompts were engineered to guide the LLM model \(Gemini\) through the specific methodological steps of Bardin’s Content Analysis\.

### A\.1Prompt: Pre\-Analysis Stage

Youareatthepre\-analysisstageandwillcarryouttheidentificationofindicesandtheselectionofindicatorsfromtheanalysisdocuments,followingLaurenceBardin’smethodology\.

Basedontheanalysisdocuments\(corpus\),youmust:

1\.IDENTIFYindices:"Traces"orsignspresentonthetext’ssurfacethatsuggestthepresenceofphenomenarelatedtothecentralquestion\.

2\.CONSTRUCTindicators:Definehowtosystematicallymeasureoridentifythepresenceorabsenceoftheseindices\.

CRITERIAFORSELECTINGINDICES

Indicesmustbe:

\-Pertinent:Relatedtothecentralquestionandhypotheses\.

\-Observable:Objectivelyidentifiableinthetext\.

\-Manifest:Explicitlypresentinthediscourse\.

\-Representative:Relevanttotheintendedinference\.

CRITERIAFORCONSTRUCTINGINDICATORS

Indicatorsmustbe:

\-Precise:Clearlydefinedandunambiguous\.

\-Reliable:Allowinganotheranalysttoreachthesameresults\.

\-Systematic:Consistentlyapplicableacrosstheentirecorpus\.

\-Fitforpurpose:Quantitativeorqualitative,asrequired\.

Contextualizationdocumentsshouldbeusedassourcesoftheoreticalconceptstounderpintheanalysis\.

Thecentralfocusofthiscontentanalysisistoanswerthequestiondrivingtheanalysistheresearch’scentralquestion\.Thisneednotnecessarilybeaquestion;itcouldbeathesisoraresearchmotivation\.

Proceedwithmethodologicalrigor\.Youranalysismustbereplicablebyanotherresearcherfollowingthesamecriteria\.

Youareatthepre\-analysisstageofthemethodologyandareextractingindicesandindicatorsfromtheanalysisdocuments,whichconsistoftranscribedinterviews\.YourresponseMUSTbeasingleblockofJSONcode,withoutMarkdownformatting,comments,orintroductorytext\.

Includepreciseandreferenceableinformation\.TheJSONmustbeanobject

containingasinglekey"indices",whichisalistofobjects\.

Additionally,eachindicatormustincludealistofreferencesindicatingwhereit

wassourcedfrom\.

ReturnONLYavalidJSONobject,withoutmarkdown,comments,oradditional

text:

TheJSONMUSTFOLLOWTHISSTRUCTUREEXACTLY:

\{\{

"indices":\[

\{\{

"name":"IndexName",

"description":"Cleardescriptionofwhatthisindexisand

whyitwaschosen,withtheoreticalgrounding",

"indicator":"PRECISEdescriptionofhowtomeasure/identifythisindex",

"references":\[

\{\{

"document":"file\_name",

"page":"2",//Always1forTXTdocuments

"quoted\_content":"Theexacttextexcerptthat

justifiesthisindex"

\}\}

\]

\}\}

\]

\}\}

THESTRUCTUREABOVEISTHEEXPECTEDFINALSTRUCTURE\.

Youwillreceiveallcontextualizationdataandmustpopulatethe

JSONwiththeindices;eachindexmustcontainanameandadescription\.The

descriptionoftheindexshouldbebriefandincludeacommentregardingthe

choiceoftheindexand/orwhatitrepresentsintheanalysis\.

CENTRALRESEARCHQUESTION:

\{0\}

CONTEXTUALIZATIONDOCUMENTSSUBMITTED\(names\):

\{1\}

ANALYSISDOCUMENTSSUBMITTED\(names\):

\{2\}

CONCRETEEXAMPLESOFINDICESANDINDICATORS\(fictitiouscontext:

Analysisofinterviewswithteachersabouttheirexperiencesduringthe

pandemic\):

\{\{

"indices":\[

\{\{

"name":"Technologicaldifficulties",

"description":"Mentionsofproblemswiththeinternet,

platforms,orcomputers",

"indicator":"Presenceofwords:’internetwentdown’,’didn’t

work’,’froze’,’couldn’taccess’"

\}\},

\{\{

"name":"Fatigueandexhaustion",

"description":"Expressionsrelatedtophysical

ormentalfatigue",

"indicator":"Presenceofwords:’tired’,’exhausted’,

’drained’,’can’ttakeitanymore’"

\}\},

\{\{

"name":"Socialisolation",

"description":"Mentionsofalackofhumancontact",

"indicator":"Presenceofwords:’alone’,’lackof

contact’,’distant’,’missthestudents’"

\}\},

\{\{

"name":"Learningnewtools",

"description":"Accountsregardinglearningtouse

technologies",

"indicator":"Presenceofexpressions:’learnedtouse’,

’discovered’,’mastered’,’gainedskills’"

\}\},

\{\{

"name":"Scheduleflexibility",

"description":"Mentionsoftime\-related

benefits",

"indicator":"Presenceofexpressions:’workingfrom

home’,’myschedule’,’moretime’,’avoidedtraffic’"

\}\}

\]

\}\}

ADDITIONALTHEORETICALFRAMEWORKREGARDINGTHECOLLECTIONOFINDICESAND

INDICATORS\(excerpttakenfromBardin’sbook\):

NOTE:THISCONTENTISEXCLUSIVELYFORCONTEXTUALIZINGTHESTAGE

YOUARECURRENTLYUNDERTAKING

Iftextsareconsideredamanifestationcontainingindicesthattheanalysiswillmakeexplicit,thepreparatoryworkinvolvesselectingthembasedonhypotheses,shouldtheybeestablishedandsystematicallyorganizingthemintoindicators\.

Asanotherexample,theindexmightbetheexplicitmentionofathemeinamessage\.Ifoneproceedsfromthepremisethatathemeholdsgreaterimportanceforthespeakerthemorefrequentlyitisrepeated\(asinquantitativesystematicanalysis\),thecorrespondingindicatorwillbethefrequencyofthatthemeeitherrelativeorabsoluteinrelationtoothers\.

Forexample:itisassumedthatemotionandanxietymanifest

throughspeechdisturbancesduringatherapeuticinterview\.Theselected

indices

\("uh\-huh,"interruptedphrases,repetition,stuttering,incoherent

sounds\.\.\.\)and

theirfrequencyofoccurrencewillserveasanindicatoroftheunderlying

emotionalstate\.

Oncetheindiceshavebeenchosen,thenextstepistoconstruct

preciseandreliable

indicators\.Fromthepre\-analysisstageonward,onemustdetermine

proceduresfor

segmentingthetextintocomparablecategorizationunitsfor

thematicanalysisandforestablishingcodingmethodsfordatarecording\.

Generally,theeffectivenessandrelevanceofthe

indicatorsareverifiedbytestingthemonafewpassagesorelementsof

thedocuments\(pre\-testanalysis\)\.

\[\.\.\.\]

Onlyindicesareretainedinanon\-frequency\-basedmanner,allowingtheanalysttoresorttoquantitativetestsforexample,theappearanceofsimilarindicesinsimilardiscourses\.

\[\.\.\.\]

Theaimofcontentanalysisistoinferknowledgeregardingtheconditionsofproduction\(or,potentially,reception\)aninference

thatreliesonindicators\(whetherquantitativeornot\)\.

\[\.\.\.\]

Letusconsideranexample:Iintendtomeasureasubject’slevelofanxietywhichtheydonotconsciouslyexpressinthemessagetheyhavetransmittedataskthatrequires,\*aposteriori\*,awrittentranscriptionofthespokenwordandvariousmanipulations\.Imightdecidetoadoptanindicatorofasemanticnature\.Forexample\(atthelevelofmeaning\),notingthefrequencyoftermsorthemesrelatedtoanxietywithinthesubject’svocabulary\.Orelse,ifitseemsvalidtome,Imightusealinguisticindicator\(thesequenceofsignifiers,thelengthof"sentences"\),oraparalinguisticone\(intonationandpauses\)\.

Ultimately,thescope,functioning,andobjectiveofcontentanalysiscanbesummarizedasfollows:currently,andingeneralterms,theterm"contentanalysis"refersto:

Asetofcommunicationanalysistechniquesaimedatobtainingthroughsystematicandobjectiveproceduresfordescribingmessagecontentindicators\(quantitativeornot\)thatallowfortheinferenceofknowledgeregardingtheconditionsofproduction/reception\(inferredvariables\)ofthosemessages\.

THEORETICALSUMMARYOFTHESTAGE:

Furthermore,duringthisphase,indicatorsaredevelopedtounderpin

thefinalinterpretation\(BARDIN,2011\)\.Ifweconsiderthatthedocuments

\(thecorpus\)areamanifestationcontaining"traces"orindicesthatthe

analysiswillmake"speak,"themissionofthisstageistoselecttheseindicesand

organizethemintoindicatorsinasystematicway\.

Indices\(traces\)arethesignsonthesurfaceofthetextor

communicationthatsuggestthepresenceofsomethingyouwishtoinvestigate\.Indicators\(measures\),

ontheotherhand,arethewayyouwillmeasureorquantify\(or

not\)thepresenceoftheseindicessothatyoucantheninferknowledge

aboutthecontext\(theinferredvariable\)\.

Imagineyourhypothesisis:Thepatient’semotionandanxiety

manifestwhentheyspeakduringatherapeuticinterview\.

Inthisexample,youtransformtheabstractphenomenon\("anxiety"\)intoa

concreteindicator\(the"frequencycount"ofincoherentsounds\),

77

allowingtheanalysistobeobjectiveandsystematic,goingbeyondwhatthe

patientexplicitlysaid\.

ConcepttobeInferred\(InferredVariable\)=Anxiety/Emotion

Hypothesis=Anxietymanifeststhroughspeechdisturbances

Indices\(TracesintheText\)=Interruptedsentences,repetition

ofwords,stuttering,incoherentsounds\(e\.g\.,"uh,""um"\)

Indicator\(SystematicMeasure\)=Thefrequencyofoccurrence

oftheseindices

### A\.2Prompt: Exploration Stage

YouareattheMATERIALEXPLORATIONstageandwillbegroupingthe\*\*unitsofanalysis\*\*into\*\*categories\*\*\.

Thecentralfocusofthiscontentanalysisistoanswerthequestiondrivingtheanalysistheresearch’scentralquestion\.Itmaynotnecessarilybeaquestion,butratherathesisoraresearchmotivation\.

Proceedwithmethodologicalrigor\.Youranalysismustbereplicablebyanotherresearcherfollowingthesamecriteria\.Yourtask:Basedontheprovidedindicesandindicators,youmust:

1\.IDENTIFYRECORDINGUNITSintheanalysisdocumentsthatcontain

thedefinedindices

2\.GROUPtheserecordingunitsintothematicCATEGORIES

3\.COUNTthefrequencyofeachcategory\(howmanyrecordingunits

belongtoit\)

COUNTINGCRITERIA\(UnitofEnumeration\)

Type:Simplefrequency\(asdescribedbyBardin\)

Recordingunit:Textsegment\(sentence/excerpt\)expressinga

completeidea

Referent:Onlystatementsmadebytheinterviewees

Countingrule:Count\+1foreachDISTINCTrecordingunitthat:

1\.Containsatleastoneoftheprovidedindices/indicators

2\.Semanticallyexpressesthecategory’scontent

3\.Isnotarepetitionofthesamethemewithinthesamecontext

4\.Hasaliteralmeaning\(disregardironyornegations\)

STRICTRULES

1\.UseONLYtheprovidedindicesandindicators\(refertothembyname

whenever\*\*citingthem\*\*,neverbytheirIDs\)

2\.EachrecordingunitmustcitetheEXACTexcerptfromthesourcedocument

3\.Categoriesmustemergefromthethematicgroupingoftheunits

4\.Donotinventorinferinformationnotpresentinthetexts

5\.Justifyeachclassificationbasedontheindicesfound

6\.AdherefaithfullytoBardin’smethod:categoriesmustbemutually

exclusive,homogeneous,pertinent,objective,andproductive

IMPORTANTNOTE:

ThereisNOTnecessarilya1:1relationshipbetweenindicesandcategories\!\-Indicesareconcreteelementsinthetext\(words,phrases,expressions\)

\-Recordingunitsaretextsegmentscontainingoneormoreindices

\-Categoriesarethematicgroupingsthatemergefrommultiplerecording

units

ReturnONLYavalidJSONobject,withoutmarkdown,comments,oradditional

text:

\{\{

"categories":\[

\{\{

"name":"CategoryName",

"definition":"Cleardescriptionofthesemanticcriteriondefiningthis

category",

"frequency":0,

"register\_units":\[

\{\{

"text":"EXACTexcerpttakenfromthedocument",

"document":"file\_name",

"page":"2",//always1forTXTdocuments

"found\_indices":\["index\_id\_1","index\_id\_2"\],

"indicator":"indicator\_id\_X",

"justification":"Explanationofwhythisunitbelongstothe

category"

\}\}

\]

\}\}

\]

\}\}

Youwillreceiveallthecontextualizationdataandmustfillinthe

JSONwiththecategories\.

ThedefinitionsofLaurenceBardin’smethodologymustbefollowedstrictly\.

THEREISNOTNECESSARILYA1:1RELATIONSHIPBETWEENCATEGORYANDINDEX\.

THEYAREMETHODOLOGICALLYDIFFERENTTHINGS\.EXAMPLEOFINDEX\-BASEDCATEGORIES\(fictitiouscontext:Analysisof

interviewswithteachersregardingtheirexperiencesduringthepandemic\):

\{\{

EXAMPLEOFPOSSIBLEINDICESANDINDICATORSFORANANALYSIS:

"indices":\[

\{\{

"id":"idx\_001",

"name":"Technologicaldifficulties",

"description":"Mentionsofproblemswiththeinternet,platforms,

orcomputers",

"indicator":"Presenceofwords:’internetwentdown’,’didn’twork’,

’froze’,’couldn’taccess’"

\}\},

\{\{

"id":"idx\_002",

"name":"Fatigueandexhaustion",

"description":"Expressionsrelatedtophysicalormentalfatigue",

"indicator":"Presenceofwords:’tired’,’exhausted’,’drained’,

’can’ttakeitanymore’"

\}\},

\{\{

"id":"idx\_003",

"name":"Socialisolation",

"description":"Mentionsofalackofhumancontact",

"indicator":"Presenceofwords:’alone’,’lackofcontact’,

’distant’,’missthestudents’"

\}\},

\{\{

"id":"idx\_004",

"name":"Learningnewtools",

"description":"Accountsoflearningtousetechnologies",

"indicator":"Presenceofexpressions:’learnedtouse’,’discovered’,

’mastered’,’gottrained’"

\}\},

\{\{

"id":"idx\_005",

"name":"Scheduleflexibility",

"description":"Mentionsoftime\-relatedbenefits",

"indicator":"Presenceofexpressions:’workfromhome’,’my

schedule’,’moretime’,’avoidedtraffic’"

\}\}

\]

\}\}

POSSIBLECATEGORIESFORTHEINDICESANDINDICATORSLISTEDABOVE:

\{\{

"categories":\[

\{\{

"name":"RemoteTeachingChallenges",

"definition":"Groupsdifficultiesandobstaclesfacedduring

onlineteaching",

"frequency":15,

"register\_units":\[

\{\{

"text":"Theinternetwentdownthreetimesduringtheclass;it

washorrible",

"document":"entrevista\_prof\_01\.pdf",

"page":"2",

"found\_indices":\["idx\_001"\],

"main\_indicator":"idx\_001",

"justification":"Reportsatechnicalissuethatdisruptedtheclass"

\}\},

\{\{

"text":"Ifeltcompletelydrained;Ididn’thavetheenergyto

preparelessons",

"document":"entrevista\_prof\_01\.pdf",

"page":"3",

"found\_indices":\["idx\_002"\],

"main\_indicator":"idx\_002",

"justification":"Expressesexhaustionrelatedtoremotework"

\}\},

\{\{

"text":"Zoomfroze,andIlostallthestudentsintheroom",

"document":"entrevista\_prof\_03\.pdf",

"page":"1",

"found\_indices":\["idx\_001"\],

"main\_indicator":"idx\_001",

"justification":"Anothertechnicalissuethatimpactedtheclass"

\}\},

\{\{

"text":"Aloneinfrontofthescreen,withoutseeinganyone’sface,

Ifeltisolated",

"document":"entrevista\_prof\_02\.pdf",

"page":"4",

"found\_indices":\["idx\_003"\],

"main\_indicator":"idx\_003",

"justification":"Reportstheemotionalimpactofisolation"

\}\}

\]

\}\},

\{\{

"name":"GainsandLessonsLearned",

"definition":"Groupspositiveaspectsandbenefitsidentified

intheexperience",

"frequency":8,

"register\_units":\[

\{\{

"text":"IlearnedtouseMoodle,GoogleClassroom,andvarious

newtools",

"document":"entrevista\_prof\_02\.pdf",

"page":"2",

"found\_indices":\["idx\_004"\],

"main\_indicator":"idx\_004",

"justification":"Reportstheacquisitionoftechnological

skills"

\}\},

\{\{

"text":"Workingfromhome,Ihadmoretimewithmyfamily",

"document":"entrevista\_prof\_04\.pdf",

"page":"3",

"found\_indices":\["idx\_005"\],

"main\_indicator":"idx\_005",

"justification":"Identifiesabenefitrelatedto

flexibility"

\}\},

\{\{

"text":"ItrainedindigitaltoolsIhadnever

usedbefore",

"document":"entrevista\_prof\_01\.pdf",

"page":"5",

"found\_indices":\["idx\_004"\],

"main\_indicator":"idx\_004",

"justification":"Anotherreportofskills

development"

\}\},

\{\{

"text":"Avoidingtwohoursoftrafficadaywasliberating",

"document":"entrevista\_prof\_03\.pdf",

"page":"2",

"found\_indices":\["idx\_005"\],

"main\_indicator":"idx\_005",

"justification":"Benefitrelatedtotimesaved"

\}\}

\]

\}\}

\]

\}\}

\}\}

INDICESANDINDICATORS\(RESULTINGFROMTHEPRE\-ANALYSISSTAGE\):

\{0\}

CENTRALRESEARCHQUESTION:

\{1\}

CONTEXTUALIZATIONDOCUMENTSSUBMITTED\(names\):

\{2\}

ANALYSISDOCUMENTSSUBMITTED\(names\):

\{3\}

ADDITIONALTHEORETICALFRAMEWORKREGARDINGTHEMATERIALEXPLORATIONAND

CODINGSTAGE\(excerptstakenfromLaurenceBardin’sbook\):

NOTE:THISCONTENTISEXCLUSIVELYFORCONTEXTUALIZINGTHESTAGE

YOUARECURRENTLYUNDERTAKING

Ifthevariouspre\-analysisoperationsareproperly

completed,theanalysisphaseitselfisnothingmorethanthe

systematicapplication

ofthedecisionsmade\.Whetherinvolvingproceduresapplied

manuallyoroperationsperformedbycomputer,theexecutionofthe

programproceedsmechanically\.Thisphase,longandtedious,consists

essentiallyofcoding,decomposition,orenumerationoperations,

basedonpreviouslyformulatedrules\.

\[\.\.\.\]

Processingthematerialmeanscodingit\.Codingcorrespondstoatransformationcarriedoutaccordingtopreciserulesoftherawtextdata,atransformationthat,throughsegmentation,aggregation,andenumeration,allowsonetoarriveatarepresentationofthecontentoritsexpression;onecapableofenlighteningtheanalystregardingthetext’scharacteristics,

whichcanserveasindices,or,asO\.R\.Holstiputsit:Codingistheprocessbywhichrawdataaresystematicallytransformedandaggregatedintounits,whichallowforanexactdescriptionofthepertinentcharacteristicsofthecontent\.

\[\.\.\.\]

Theorganizationofcodinginvolvesthreechoices\(inthecaseofa

quantitativeandcategoricalanalysis\):

\*Segmentation:choiceofunits;

\*Enumeration:choiceofcountingrules;\(NOTE:THISSTAGEHASALREADY

BEENCOMPLETEDANDWILLBEPROVIDED\)

\*Classificationandaggregation:choiceofcategories\.\(NOTE:WEARE

CURRENTLYATTHISSTAGE\)

\[\.\.\.\]

RECORDINGANDCONTEXTUNITS

Whichelementsofthetextshouldbetakenintoaccount?Howshouldthetextbesegmentedintocompleteunits?Thechoiceofrecordingandcontextunitsmustbemadeappropriately\(appropriatenessregardingthematerial’scharacteristicsandtheanalysisobjectives\)\.

a\)Therecordingunit:Thisisthecodedunitofmeaning;itcorrespondstothecontentsegmenttreatedasthebasicunitforthepurposesofcategorizationandfrequencycounting\.Therecordingunitcanvarywidelyinnatureandsize\.

Acertainambiguityprevailsregardingthecriteriafordistinguishingrecordingunits\.Indeed,certainsegmentationsarecarriedoutatasemanticlevelforexample,the"theme"whileothersaremadeatanapparentlylinguisticlevel,suchasthe"word"orthe"sentence\."

Thisservesasacritiqueofdisciplineswhosescientificandrigorousnatureismoreevident\.

Infact,thesegmentationcriterionincontentanalysisisalwayssemanticinnature,eventhoughtheremaysometimesbeacorrespondencewithformalunits\(examples:wordandkeyword/thematicword;sentenceandsignifyingunit\)\.

Bywayofillustration,thefollowingcanbecitedamongthemostcommonlyusedrecordingunits:

\*Theword:whileitistruethatthe"word"lacksaprecisedefinitioninlinguistics,forlanguageusersitcorrespondstosomethingtangible\.However,linguisticprecisioncanbeintroducedifrelevant\.

Allwordsinthetextmaybetakenintoaccount,oronemayretainonlykeywordsorthematicwords\(\*symbols\*inEnglish\);adistinctioncanalsobemadebetweencontentwordsandfunctionwords;

orananalysisofaspecificwordcategorymaybeperformed:nouns,adjectives,verbs,adverbs\(\.\.\.\)inordertoestablishratios

\[\.\.\.\]

Thecontextunit:

Thecontextunitservesastheunitofcomprehensionforcodingtherecordingunit;itcorrespondstothesegmentofthemessagewhosedimensions\(largerthanthoseoftherecordingunit\)areoptimalforgraspingtheexactmeaningoftherecordingunit\.

Thismight,forexample,bethesentenceforawordandtheparagraphforatheme\.

Indeed,inmanycases,itbecomesnecessarytorefer\(consciously\)totheimmediateorbroadercontextoftheunitbeingrecorded\.Ifmultiplecodersareworkingonthesamecorpus,prioragreementisessential\.

Forinstance,whenanalyzingpoliticalmessages,wordssuchasliberty,order,progress,democracy,andsocietyrequirecontexttobeunderstoodintheirtruesense\.

Referencetocontextiscrucialforbothevaluativeanalysisandcontingencyanalysis\.Resultsareliabletovarysignificantlydependingonthedimensionsofthecontextunit\.Theintensityandscopeofaunitmayappearmoreorlesspronounced,dependingonthedimensionsofthechosencontextunit\.

Regardingco\-occurrences,itisevidentthattheirnumberincreaseswiththedimensionsofthecontextunit:forexample,itisunlikelythatsimilarthemeswouldbefoundwithinasingleparagraphorafewminutesofarecording,

whereastheprobabilityincreasesinamulti\-pagetextoraonehourbroadcast\.Generally,thelargerthecontextunit,themorestronglyattitudesorvaluesemergeinanevaluativeanalysis,orthemorenumeroustheco\-occurrencesinacontingencyanalysis\.

Determiningthedimensionsofthecontextunitisgovernedby

twocriteria:costandrelevance\.Clearly,aunitofbroadcontextrequiresamoreextensivere\-examinationofthematerial\.Ontheotherhand,thereisanoptimaldimensionintermsofmeaning:ifthecontextunitistoosmallortoolarge,itisnolongersuitable;here,too,boththetypeofmaterialandthetheoreticalframeworkaredecisivefactors\.

Inanycase,itispossibletotesttherecordingandcontextunits

onsmallsamplestoensurethatweareworkingwiththe

mostappropriateinstruments

\[\.\.\.\]

138CONTENTANALYSIS

2\.ENUMERATIONRULES

Adistinctionmustbemadebetweentherecordingunitwhatiscountedandtheenumerationrulethemethodofcounting\.

Letusconsiderthefollowingexample:wehaveacompleted"text"

inwhichtheidentificationandsegmentationprocessesyieldedthefollowing

elementsorrecordingunits\(words,themes,orotherunits\):

a,d,a,e,a,b\.Giventhatthereferencelistestablishedbasedonasetof"texts"oraccordingtoastandardisa,b,c,d,e,f,itispossibletousevarioustypesofenumerations:

Frequency\(NOTE:THEMEASURETOBEUSEDINOURCASE\):thisisgenerallythe

mostwidelyusedmeasure\.Itcorrespondstothe

followingpostulate\(validinsomecasesbutnotothers\):theimportanceof

arecordingunitincreaseswithitsfrequencyofoccurrence\.Inour

example,thefrequencyofeachelementis:

a=3;

b=1;

c=0;

d=1;

e=1;

f=0\.

Afrequencymeasureinwhichalloccurrencescarrythesameweight

postulatesthatallelementsareofequalimportance\.

Thechoiceofasimplefrequencymeasureshouldnotbeautomatic\.Wemust

rememberthatitrestsonthefollowingimplicitassumption:theoccurrenceofanitemofmeaningorexpressionbecomesincreasinglysignificantinrelationtotheintendeddescriptionorinterpretationoftherealityunderstudythemorefrequentlyitrecurs\.Quantitative

regularityofoccurrenceis,therefore,whatisdeemedsignificant\.This

presupposesthatallitemshavethesamevalue,whichisnot

alwaysthecase\.

\[\.\.\.\]

Categorizationisanoperationinvolvingtheclassificationoftheconstituent

elementsofasetthroughdifferentiationandsubsequentregrouping

accordingtogenre\(analogy\),basedon

predefinedcriteria\.

Categoriesareheadingsorclassesthatgatheragroupofelements\(recordingunits,inthecaseofcontentanalysis\)underagenerictitle\-agroupingperformedbasedonthecommoncharacteristicsoftheseelements\.

Thecategorizationcriterionmaybesemantic\(thematiccategories:forexample,allthemessignifyinganxietyaregroupedunderthecategory"anxiety,"whilethosesignifyingrelaxationaregroupedundertheconceptualheading"relaxation"\),syntactic\(verbs,adjectives\),

lexical\(classificationofwordsaccordingtotheirmeaning,pairingsynonymsandcloselyrelatedmeanings\),andexpressive\(forexample,categoriesthat

classifyvariouslanguagedisorders\)\.

\[\.\.\.\]

CONTENTANALYSIS

Taxonomicactivityisaverycommonoperationinvolvingthedistribution

ofobjectsintocategories\.If,beforeplacingarecordontheturntable,

weaskourselveswhetherwewanttolistentoBach,Ravel,orBoulez,

wearenotusingthesamecriterionthatguidesourchoicesifweaskourselves

whetherwewishtoheartheviolin,organ,orpiano\.Thecategorizationcriterion

isnotthesame\(composerversusinstrument\)\.Wearenotemphasizingthesameaspectofreality\.Furthermore,thecriterionweemployismoreorlesssuitedtotherealitypresentedtous\.Itispossiblethatourtwodesiresmightconverge,therebyrefiningthechoicewemake\(aspecificinstrumentandaspecificcomposer\)\.Similarly,incontentanalysis,amessagemaybesubjectedtooneorseveraldimensionsofanalysis\.

Classifyingelementsintocategoriesrequiresinvestigatingwhateachonehasincommonwiththeothers\.Whatenablestheirgroupingisthecommonelementsharedamongthem\.However,othercriteriamightfocusondifferentaspectsofsimilarity,potentiallyalteringthepreviousdistributionsignificantly\.

Categorizationisastructuralistprocesscomprisingtwo

stages:

\*inventory:isolatingtheelements;

\*classification:distributingelementsand,consequently,seekingorimposing

acertainorganizationonthemessages\.

\[\.\.\.\]

EXAMPLESOFCATEGORYSETS

Whileinmostcasesitisnecessarytocreateacategorygridforeachnewanalysis,previousstudiescanserveasinspirationfortheanalyst\.Forthisreason,wewillcitesomeexamplesofcategorysetsthathavealreadybeenused\.

CATEGORIZATION

a\)Valueanalysis

Example1:

ShortlyafterWorldWarII,Whitespecializedinvalueanalysis\.

HefirstanalyzesRichardWright’sautobiography,\*BlackBoy\*\(1947\);

hethenanalyzesthepropagandastylesofHitlerandRoosevelt\(1949\)

and,later,thespeechesofKennedyandKhrushchev\(1967\)\.

Wepresentoneofhisanalysisgrids\.A/Physiologicalvalues

1\.Food/Nutrition

2\.Sex

3\.Rest

4\.Health

5\.Safety/Security

6\.Comfort

D/Valuesexpressingfear

\(emotionalsecurity\)

E/Valuesofplayandjoy

1\.Newexperience

2\.Excitement,emotion

3\.Beauty

4\.Humor

5\.Creativeself\-expression

F/Practicalvalues

1\.Practicality

2\.Possession

3\.Work

G/Cognitivevalues

1\.Knowledge

H/Miscellaneous

1\.Happiness

2\.Valueingeneral

B/Socialvalues

1\.Sexuallove

2\.Familylove

3\.Friendship

C/Ego\-relatedvalues

1\.Independence

2\.Achievement/Fulfillment

3\.Recognition

4\.Self\-esteem/Self\-love

5\.Domination

6\.Aggression

Example2:

V\.Isambert\-Jamatidemonstratedtheevolutionofvaluesadvocatedbythe

schoolsystembetween1860and1965,basedontheanalysisofasampleofprize\-givingspeecheseliveredbyvariousspeakersdirectlyorindirectlyinvolvedinsecondaryeducation,producedregularlyduring2\.R\.K\.White,Value\-analysis:thenatureanduseofthemethod,GlenGardiner,N\.J\.,LibertarianPress,1951\.CitedbyHolsti,op\.cit\.

3\.IsambertJamati,Crisesdelasociete,crisesdel’enseignement,P\.U\.F\.,1970\.

CONTENTANALYSIS

thatperiodandeasilyaccessible\.Theseprize\-givingspeechesservedas

sourcematerialforacomprehensivestudyontheschool’s"referencemorality,"regardingtheendsandthemeanstoachievethoseendspursuedbytheschoolsystem,aswellastheobjectsofintellectualknowledgetobepromoted,etc\.

Asetoffivecategoriesandsubcategoriesservedasthebasisfortheanalysis\.

\*Thechangesthattheteachingofschoolsubjectsshouldbringaboutin

students:

\*Participationinsupremevalues\.

\*Individualself\-improvementsoughtbythestudent\.

\*Exerciseofoperationalmechanisms\.

\*Theobjectsofknowledge:

\*Peopleofthepastandtheirworks\.

\*Contemporarypeople\.

\*Humananduniversalnature\.

\*Nature\.

\*Theobjectsofmoraleducation:

\*Loyaltytothesecularnationaluniversity,

\*Loyaltytotheinstitution\.

\*Withdrawalfromtheworldasafavorableconditionforeducation,

\*Educationalvalueofdiscipline\.

\*Roleofpeersincharacterformation\.

\*Considerationofindividualdifferencesamongstudents\.

\*Utilizationofplayfultendencies\.

\*Teachers’moralexample\.

\*Teachers’voluntaryinfluence\.

\*Theinstitutionaldefinition:

\*Itisbeneficialforthecoredefinitionofsecondaryeducationtochange

sothatitadaptstosocialshifts\.

\*Secondaryschoolingshouldbeoflongduration\.

\*Secondaryeducationshouldsufficeforstudents,withouttheneed

forthemtocontinuetheirstudies\.

\*Lyceesshouldnotservetopreparestudentsfortheirfuturecareers\.

\*Thetargetaudienceisthesocialelite\.

CATEGORIZATION

\*Referencevalues:

\*Individualmoralitybasedonperfectionorthecategoricalimperative\.

\*Individualmoralitywithahedonistictendency,orofthe"mentalhygiene"type:

\*Individualmoralitybasedonsolidarity\.

\*Exhortationtowork\.

\*Exaltationofprogress\.

\*Exaltationofyouth\.

\*Exaltationofthefamily\.

\*Exaltationofthehomeland\.

\*Exaltationofpeaceandinternationalunderstanding\.

ThefinalconclusionofthisstudydemonstratesthatchangesinFrenchsociety

arereflectedintheobjectivesproposedbyeducationalsystems,and

thatsocietalcrisesandeducationalcrisesappearsynchronized\.Theobjectivesoftheschoolinstitutionevolve\.Thus,itispossibletodividetheperiodsaccordingtodominantvalues:

1\)1860\-1870:supremevaluesandintegrationintotheelite\.

2\)1876\-1885:integrationintotheeliteandtransformationoftheworld\.

3\)1896\-1905:transformationoftheworldandsecularenthusiasm\.

4\)1906\-1930:cultureasafreegood\.

5\)1931\-1940:learningtolearn\.

6\)1946\-1960:secondaryeducationdefendsitself:returntoaestheticism\.

7\)1961\-1965:crisisofobjectives\.

Fromatechnicalstandpoint,theanalyseswereessentiallythematic,yetalwaysrefinedbyprecautionssuchastheweightingofthemes,thedivisionintoprimaryandsecondarythemes,anevaluativeapproach\(favorabletextvs\.neutraltext\),andtheuseofgenderrelationspecificallya"dominancecoefficient":

b\)Analysisofendsandmeans

Example1:

ThisisananalysisofaffectiveandrationalobjectivesconductedbyB\.BerelsonandP\.Salterregardingpopularfictionmagazines\.

Twocategorysystemswereused:

4\.AndthencameMay’68\!5B\.BerelsonandP\.Salter,"MajorityandminorityAmericans:ananalysisofmagazinefiction,"\*PublicOpinionQuarterly\*,1946\.

CONTENTANALYSIS

A/Intentionsofthe"heart"

1\.Romanticlove

2\.Establishedmarriage

3\.Idealism

B/Intentionsofthe"head"

1\.Solvingconcreteproblems

2\.Personalprogress

3\.Moneyandmaterialgoods

4\.Affectionandemotionalsecurity

5\.Economicandsocialsecurity

6\.Poweranddomination

7\.Patriotism

8\.Adventure

9\.Justice

10\.Independence

Example2:

Thisstudyanalyzesthegoalsandpossibilitiesforsuccessofferedtochildrenintelevisionprograms,relatingthemtotheadvocatedmeans\.

A/Goalcategories

1\.Property\(materialsuccess\)

2\.Self\-preservation\(includingdesireforthestatusquo\)

3\.Affection

4\.Sentiment

5\.Powerandprestige

6\.Psychologicalobjectives\(includingviolenceandeducation\)

7\.Others

B/Methodcategories

1\.Legal

2\.Non\-legal\(withoutinjuryordamage\)

3\.Economic

4\.Violence

5\.Organization,negotiation,andcompromise

6\.Evasion,flight\(attempttoavoidfactsinherenttoachievingthe

goal,forgettingtheobjective,etc\.\)

7\.Chance

8\.Others

6\.O\.N\.Larson,L\.N\.Gray,andJ\.G\.Fortis,"Goalsandgoal\-achievement

methodsintelevision

content:modelforanomie?"in\*SocialInquiry\*,33,1963\.