Analyzer

The Analyzer is the module responsible for performing semantic analysis on documents and assistant conversations. It allows CogSol to transform unstructured language into structured insights through configurable analytical models. It can analyze both user conversations or already processed documents on the implementation portal.

What is Semantic Analysis?

Semantic analysis is the process of interpreting and categorizing text to extract meaning, patterns, and quality indicators. In the context of CogSol, the Analyzer uses semantic analysis to evaluate how information is expressed, whether it meets certain criteria, and how it compares across datasets or assistants.

Purpose of the Analyzer

The primary purposes of the Analyzer are to:

• Enable consistent and explainable evaluation of text-based data.
• Support quality monitoring for assistants and document corpora.
• Identify patterns, strengths, and anomalies in large-scale textual interactions.
• Provide structured outputs (dimensions, features, and alerts) that can be used for decision-making or further automation.

Key Components

Analyzer Configuration

An Analyzer defines what aspects of a text should be analyzed. Each Analyzer includes:

• General Configuration: basic metadata (name, description, type: documents or chats).
• Dimensions: conceptual axes of analysis (e.g., accuracy, tone, coherence).
• Features: measurable or observable elements within each dimension.
• Alerts: logical conditions that trigger notifications or warnings based on results.
• Summary (optional): a synthesized interpretation of the overall findings.

These configurations are reusable and can be applied to different datasets or assistants.

Dimensions

Dimensions define what is being analyzed. They group related features under broader analytical themes such as:

• Linguistic Quality: evaluates grammar, clarity, and consistency.
• Informational Accuracy: checks for factual correctness and citation completeness.
• Interaction Quality: assesses engagement, tone, and empathy in conversations.
• Policy Compliance: detects adherence to internal or regulatory guidelines.

Each dimension provides a structured view of the text from a specific analytical perspective.

Features

Features define how the analysis is measured. They are the smallest observable units, describing specific behaviors or properties within a dimension. Examples may include:

• “Contains reference link” (Boolean)
• “Tone: Positive / Neutral / Negative” (Categorical)
• “Response completeness” (Scale)

Features allow the Analyzer to translate semantic meaning into measurable results.

Alerts

Alerts are automatic conditions used to highlight relevant findings or deviations. They can be configured to detect situations like:

• Missing required information.
• Inconsistent factual statements.
• Repeated communication issues (e.g., fallback loops).

Alerts provide actionable signals to users or monitoring systems, allowing quick detection of quality or compliance issues.

Summary

The Summary is an optional synthesis that aggregates the findings of an Analyzer execution. It provides a concise natural-language interpretation of the overall analysis, summarizing the most important patterns, alerts, and outcomes.

Analyzer Execution

Once configured, an Analyzer can be executed over documents already uploaded to the platform or over assistant conversations (chats).

• For document analysis, the Analyzer processes all documents within a selected node (topic) and its children.
• For chat analysis, the Analyzer evaluates conversations associated with a specific assistant.

Document analysis can be triggered from the topics detail screen using the Analyze button. This will show the available already conmfigured analyzers and from there, run the analysis for the selected node (topic) and its children.

Chat analysis can be performed through the implementation portal by clicking the actions button over an assistant with the following analysis options:

• Rerun with existing info: re-executes the analysis using previously collected data without fetching new information.
• Analyze all chats of the assistant: performs analysis on the complete conversation history for a specific assistant.
• Add new chats to the analysis from a date range: includes additional conversations from a specified time period to the existing analysis.
• Add new chats to the analysis from a specific date: incorporates new conversations starting from a particular date.

The execution process for both methods follows these conceptual steps:

1. Data Collection: retrieves and normalizes the input dataset.
2. Semantic Interpretation: processes text through CogSol’s Semantic Engine.
3. Mapping: aligns semantic results with defined dimensions and features.
4. Alert Evaluation: checks for conditions that trigger alerts.
5. Aggregation: compiles all results into an analytical summary.

Executions can be triggered manually or automatically, and results are stored for comparison and trend analysis.

Outputs

An Analyzer produces structured results that can be explored through the platform:

• Dimension-level metrics: how each analytical aspect performed.
• Feature results: detailed outcomes for each evaluated property.
• Alerts: list of triggered conditions with their severity levels.
• Summary: global interpretation of findings.

These outputs feed other modules (e.g., Dashboards, Cognitive Layer) or can be exported for external monitoring.

Reporting and Visualization

Analyzer reports allow users to explore results interactively through a comprehensive dashboard interface. The visualization includes several key components:

Filters

The reporting interface provides powerful filtering capabilities to refine analysis results. The user may select several filters to see the results of a custom query:

• Characteristics: filter by specific features identified during analysis.
• Metadata: filter by metadata properties and values associated with the analyzed content.
• Alerts: filter results by triggered alerts to focus on specific issues.
• Date: filter by date range to analyze specific time periods.
• Expert mode: only shows the query input field so the user can run custom queries regarding its own interests combining all the above.

Metrics and Alerts Summary

The dashboard displays a high-level overview of the analysis execution:

• Documents Analyzed: total count of documents processed in the execution.
• Execution Status: current state of the analysis (e.g., READY, PENDING, COMPLETED, ERROR).
• Total Features: number of features evaluated across all dimensions.
• Total Summaries: count of generated summaries for the analysis.
• Total Dimensions: number of analytical dimensions configured.
• Total Alerts: count of alerts triggered during the analysis.

Dimensions Visualization

Each dimension is presented with detailed visualizations:

• Frequency charts: bar graphs showing the distribution of feature values across analyzed documents.
• Total instances: count of occurrences for each dimension.
• Feature breakdown: individual features are displayed on the x-axis, with frequency on the y-axis.
• Interactive controls: options to view detailed data or export dimension results.

Document-level Results

NOTE: At this point, talking about documents is referring for processed document uploaded or chat. If chats were analyzed, the metadata will have the chat id added alongside other specific metadata it already has.

A detailed table shows analysis results for individual documents:

• Date: when the document was analyzed.
• Summary: optional summary for each document in natural language.
• Characteristics: tagged features identified in the document.
• Text: view the full document text.
• Metadata: associated metadata properties for each document.
• Alerts: any alerts triggered for specific documents.

The reporting interface enables users to drill down from high-level metrics to individual document analysis, facilitating both strategic insights and detailed quality assessment.

Benefits of the Analyzer

• Consistency: applies the same analytical logic across all evaluations.
• Explainability: makes language evaluation transparent and interpretable.
• Reusability: analyzers can be defined once and applied to multiple contexts.
• Automation: reduces the need for manual review while preserving contextual understanding.
• Integration: results can feed dashboards, reports, or trigger alerts automatically.

Limitations

While the Analyzer provides structured, explainable insights, it does not replace human interpretation entirely.

Some nuanced aspects — such as emotional subtleties, intent misalignment, or domain-specific context — may still require expert review.

However, as part of the CogSol ecosystem, the Analyzer acts as a scalable foundation for continuous semantic monitoring and improvement.