Automated Text Analysis Based on Skip-Gram Model for Food Evaluation in Predicting Consumer Acceptance

Area of Science:

• Food Science
• Natural Language Processing
• Consumer Behavior Analysis

Background:

• Traditional food sensory evaluation for consumer acceptance often requires extensive taste descriptive word lexicons.
• Existing methods struggle to analyze large numbers of evaluators, limiting predictions of overall consumer acceptance.
• Analyzing unstructured online reviews presents a scalable alternative to traditional sensory panels.

Purpose of the Study:

• To develop and validate an automated text analysis method for evaluating food taste and smell characteristics.
• To assess the feasibility of using consumer online reviews (from social networking services - SNS) to predict food acceptance.
• To compare the effectiveness of this automated method against traditional consumer preference taste evaluations.

Main Methods:

• Collected consumer reviews for two jjampong ramen types from SNS to avoid building a sensory word lexicon.
• Trained a word embedding model using the acquired reviews to convert text into meaningful vectors.
• Performed inference on word vectors to evaluate the taste and smell profiles of the two jjampong ramen products.
• Compared the automated method's results with a traditional consumer preference taste evaluation for validation.

Main Results:

• The automated text analysis method successfully evaluated taste and smell attributes of jjampong ramen based on online reviews.
• Vector representations of words from reviews enabled nuanced analysis of sensory characteristics.
• The findings demonstrated a high degree of reliability when compared to direct consumer taste preference data.
• The proposed method offers a scalable and efficient alternative for food sensory evaluation.

Conclusions:

• Automated text analysis of online reviews provides a reliable and scalable method for food sensory evaluation.
• This approach overcomes limitations of traditional methods by leveraging large volumes of consumer-generated data.
• The study validates the use of word embedding models for predicting consumer acceptance based on textual feedback.
• The developed technique is applicable to various food products beyond the studied jjampong ramen types.