SatisFactory’s new AI-powered semantic analysis model

Semantic analysis for analyzing customer reviews and conversations

Semantic analysis(also known as "open-ended response analysis") is an advanced natural language processing (NLP) method that can be applied in the context of feedback management to comments collected from various sources. Its value lies in its ability to measure, group, and prioritize the topics mentioned by respondents, whether they come from satisfaction surveys, online review sites, or social media.

At the heart of this approach lies the semantic classification scheme(or classification framework), which structures the analysis and is defined in advance. It establishes the themes, subthemes, and concepts expected in the verbatim transcripts, providing a coherent framework for transforming a large volume of raw comments into clear, directly actionable insights. This nuanced analysis makes it possible, in particular, to categorize a single comment under multiple themes, depending on the breadth of topics it addresses.

The semantic analysis offered by SatisFactory therefore makes it possible to:

• Classify each comment into one or more themes and subthemes, based on the identified concepts
• Prioritize identified customer expectations based on their importance and impact
• Identify the sentiment (positive, negative, neutral) associated with each concept discussed in the comments
• Highlight persistent pain points, weak signals, and emerging trends
• Provide strategic and operational teams with a clear, structured, and actionable overview to guide their decisions

Below is a comparison between SatisFactory’s old and new semantic analysis models:

Analytical model	Old Sector-Specific Semantic Model Based on Keywords	New AI-powered Semantic Model
Type of analysis	Keyword Analysis Analysis of the sentiments associated with each concept (positive, negative, or neutral)	Analysis of Terms and Concepts Sentiment analysis (positive, negative, or neutral)
How it works	Automatic classification based on keywords that have been manually assigned to each topic on the platform	Automatic classification using AI, based on the description associated with each concept
Activation	Application to the CSM Must be enabled in the program settings on the platform Final activation by Data Science SatisFactory once the configurations have been completed and validated	Application to the CSM Must be enabled in the program settings on the platform Final activation by Data Science SatisFactory once the configurations have been completed and validated
Settings	Manual configuration, which is generally inflexible and time-consuming Application to the CSM Formulation of themes and subthemes, as well as associated keywords Configuring themes and subthemes on the platform Manually add each keyword to be analyzed to the platform Implementation of semantic keyword analysis of the comment history available on the account	Semi-automatic configuration that is generally scalable and fast Application to the CSM Formulation of themes and subthemes, along with their associated descriptions Configuring themes and subthemes on the platform No further action is required on the platform Implementation of AI-powered semantic analysis of the comment history available on the account

Sentiment analysis, a key component of our semantic analysis, automatically assigns a polarity (or tone) to each topic mentioned in a comment:

• Positive:The comment expresses a favorable opinion, satisfaction, or a positive emotion regarding the subject
• Negative:The comment expresses dissatisfaction, criticism, or a pain point regarding the subject
• Neutral:The comment is factual, incomplete, or does not express any strong emotion

The security of the new semantic engine

When it comes to security, the new AI-powered semantic analysis model is based on Google Gemini.

When using the artificial intelligence features of the SatisFactory platform, we guaranteethe utmost confidentiality of your data. Our commitments are based on three pillars:

1. Systematic anonymization of verbatim data:Before any comments are submitted to the artificial intelligence system, the prompts are automatically cleaned and anonymized
2. Local data sovereignty:Data processing using generative artificial intelligence is carried out exclusively on French servers hosted in Paris
3. Limited use of data by our partner:Google formally commits to never using your data to train its models and to never sharing it with any third parties

To learn more about data security on the platform, please visit the SatisFactory Trust Page.

Calculating semantic tone

Introduction to Semantic Tone

Semantic analysis determinesthe sentiment of commentsby categorizing them into three tones:

• Positive tone
• Negative tone
• Neutral tone

This classification providestwo levels of sentiment analysis.

1. First,the comment is classified as a whole as positive, negative, or neutral.
2. Next, in addition to your semantic classification scheme and the overall sentiment analysis of the comment,each sub-theme identified in the comment is assigned a specific sentiment (positive, negative, or neutral).

This comprehensive process addresses all analytical needs, providing both an overall assessment of the feedback and a detailed understanding of the sentiment associated with each topic mentioned in the respondent’s response.

Explanation of how semantic tone is calculated

When semantic tone analysis is triggered for a comment, the following steps take place:

1. Retrieving the respondent's full response
2. Analysis of each rating provided by the respondent (key indicators and satisfaction items/attributes).
3. Identification and categorization of themes and subthemes in the commentary (according to the classification scheme defined on the platform)
4. Assigning a tone (positive, negative, or neutral) to each subtheme identified in the comment
5. Overall classification of the tone of the comment (positive, negative, or neutral) based on the balance of tones across each identified subtopic

⚠️ Comments that are too short (fewer than 3 words) are excluded from the sentiment analysis and areclassified as neutral by default.

💭 Todetermine the overall tone of a comment:

• If the number of negative subthemes is greater than or equal to the number of positive subthemes,the overall tone is classified as negative.

• Similarly, if the sentiments in the identified subthemes are evenly split (with an equal number of negative and positive sentiments as there are neutral ones),the strong sentiment (positive or negative) takes precedence over the neutral sentiment.

Enable the new semantic analysis engine

The new AI-powered semantic model is deployed very quickly, managed by a dedicated Customer Success Manager, and developed collaboratively to ensure it is tailored to your specific needs.

Within a few hours of activating the semantic model on the platform, semantic analysis becomes available to all users.This advanced analysis, tailored to your specific context, is also seamlessly integrated into many other SatisFactory features to enhance your analyses and facilitate decision-making.

Are you interested in learning more about our AI-powered semantic solution?

Contact us