Review Scoring Model Using Textual and Metadata Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems face challenges in evaluating the quality of user-generated reviews, particularly those without ratings, and aggregating scores from multiple review websites, as users often rate reviews based on agreement rather than quality, and integrating scores from different platforms is difficult.
Innovation Solution
A review scoring model is trained using textual features of reviews and metadata from authors, including social networking data, to predict scores, with constraints such as author consistency, trust consistency, and link consistency applied to adjust predicted scores.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If user-generated reviews are allowed on online portals, then user feedback and product insights are improved, but review quality varies significantly with misspellings, profanity, and spam
Solution Approach 1:
The patent replaces manual review quality assessment by users with an automated machine learning model that analyzes review text and metadata to generate quality scores. This substitution of mechanical human judgment with automated computational analysis resolves the contradiction by providing consistent, scalable quality evaluation without requiring user intervention, thereby maintaining information quality while ensuring reliability through algorithmic consistency
Solution Approach 2:
The system enables reviews to self-evaluate their own quality through the automated scoring model that analyzes each review's textual features, metadata, and author characteristics. This self-service approach allows reviews to be automatically assessed and ranked without external human intervention, improving both the efficiency of quality control and the consistency of evaluation across all reviews
2Productivity
If users rate reviews based on agreement with opinions, then user engagement is improved, but scoring accuracy deteriorates as ratings reflect agreement rather than quality
Solution Approach 1:
The patent extracts the quality assessment function from user opinions and agreement ratings. Instead of relying on users to score reviews based on their subjective agreement with the content, the system extracts objective quality signals from textual features, metadata, and author characteristics. This separation of engagement (user rating) from quality assessment (automated scoring) resolves the contradiction by maintaining high user engagement while achieving accurate quality measurement through automated analysis
Solution Approach 2:
The automated machine learning model serves as an intermediary between user engagement and review quality assessment. Users can continue to engage with reviews by expressing agreement or disagreement, while the intermediary system independently evaluates quality based on objective textual and metadata features. This intermediary layer resolves the contradiction by decoupling engagement metrics from quality scoring, allowing both to operate independently with their respective optimization goals
3Quantity of substance
If reviews are aggregated from multiple review websites, then review coverage is improved, but score integration becomes difficult due to different scoring systems
Solution Approach 1:
The patent creates a universal quality scoring system that can evaluate reviews from multiple different websites despite their varying native scoring systems. The machine learning model processes diverse review formats, metadata structures, and author profiles through a unified analysis framework, generating standardized quality scores that enable cross-platform aggregation. This universal approach resolves the contradiction by providing a common evaluation language that works across different review platforms without requiring complex platform-specific integration logic
Solution Approach 2:
The system transforms reviews from different websites with varying scoring parameters into a unified quality assessment framework. By changing the parameter representation from platform-specific scores to standardized quality dimensions (textual features, metadata quality, author characteristics), the system enables meaningful aggregation across sources. This parameter transformation resolves the contradiction by converting heterogeneous scoring systems into comparable quality metrics that can be integrated across multiple review websites
4Ease of manufacture
If traditional review filtering is used, then simple implementation is maintained, but inability to handle unrated reviews and new products limits effectiveness
Solution Approach 1:
The patent applies preliminary quality assessment to reviews before they receive user ratings or for new products without existing review histories. The machine learning model proactively evaluates incoming reviews based on their textual features, metadata, and author characteristics, assigning quality scores in advance. This preliminary action resolves the contradiction by providing quality assessment capability for unrated and new reviews while maintaining a relatively simple implementation through automated analysis rather than complex manual review processes
Data Source
AI summary
User generated reviews and scores associated with the reviews may be used to train a review scoring model with textual features of the reviews. The review scoring model may be used to predict scores for newly received reviews. One or more constraints based on social networking application data associated with an author of a review may be used to adjust the predicted score of the review.