Interest Classification Model for Social Media Recommendation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The rapid growth of information on social media platforms leads to a junk information overflow, with existing recommendation technologies, such as collaborative filtering and spectrum analysis, being ineffective due to the vast amount of untagged information and the temporal nature of social media content, resulting in poor recommendation effectiveness.

Innovation Solution

An information recommendation method and apparatus that involves an off-line procedure to determine a user's point of interest from historical operation data, select relevant information from quality users, and train an interest classification model, which is then used in an on-line procedure to recommend real-time information that aligns with the user's interests.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If information recommendation systems use collaborative filtering or spectrum analysis based on information tags, then recommendation coverage can be achieved, but recommendation effectiveness deteriorates due to the vast amount of untagged information and temporal nature of social media content

Engineering Contradiction:

Improverecommendation coverageVSAvoidrecommendation effectiveness

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The patent replaces manual information tagging (mechanical process) with automated interest classification models that automatically categorize social media information based on user historical operations. This substitution enables the system to handle untagged information effectively while maintaining recommendation quality, resolving the contradiction between coverage and effectiveness.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system enables users to automatically generate their own interest profiles through their historical operations (posts, likes, comments) without requiring manual input or tagging. The interest classification model self-adapts to user preferences by analyzing these operations, achieving both broad coverage and personalized effectiveness simultaneously.

Inventive Principle:

Principle #25Self-service

2Measurement precision

If manual annotation is used to create training samples for recommendation systems, then information accuracy can be improved, but system complexity and time consumption increase significantly

Engineering Contradiction:

Improveinformation accuracyVSAvoidsystem complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The system automatically generates training samples by collecting and analyzing user historical operations (posts, likes, comments, reposts) to create interest classification data. This self-service approach eliminates manual annotation while maintaining high accuracy, as the training data is derived directly from actual user behavior patterns.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent copies user behavior patterns from historical operations to create training samples that reflect actual user interests. By replicating real user interaction data as training data, the system achieves high accuracy without manual intervention, reducing both complexity and time consumption.

Inventive Principle:

Principle #26Copying

3Measurement precision

If all social media information is tagged for recommendation, then recommendation precision can be improved, but information loss and user privacy concerns increase

Engineering Contradiction:

Improverecommendation precisionVSAvoiduser privacy

Core Design Contradiction:

Measurement precisionVSLoss of information

Solution Approach 1:

The system extracts only the necessary features from user historical operations (such as interaction patterns, content types, and engagement metrics) to build interest profiles, rather than storing or processing all user information. This extraction approach maintains recommendation precision while minimizing privacy intrusion and information loss.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent applies different processing levels to different types of information: sensitive personal data is minimized or anonymized, while only essential behavioral patterns are retained for interest classification. This local quality approach ensures precision where needed while protecting privacy where sensitive.

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS10810499B2Method and apparatus for recommending social media information

Publication Date: 2020.10.20 HUAWEI TECH CO LTD

AI summary

The present invention provides an information recommendation method and apparatus in a social media. An off-line procedure includes: determining a point of interest of a target user; selecting information related to the point of interest as an annotated corpus; and training an interest classification model of the target user by using the annotated corpus as a training sample. An on-line procedure includes: inputting to-be-recommended information to the interest classification model of the target user, so as to determine whether the to-be-recommended information tallies with an interest of the target user; and if the to-be-recommended information tallies with the interest of the target user, recommending the to-be-recommended information to the target user. According to the present invention, an effect of information recommendation in a social media can be improved.