Social Network Profile Content Filtering via Dictionary Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Online social networks face challenges in efficiently filtering out disallowed content, such as obscenity and pornography, which can be offensive to users and pose a risk to children, requiring an effective mechanism for content identification and removal.
Innovation Solution
A system that uses 'bad-word' and 'good-word' dictionaries generated by human review to score member profiles, with logistic regression to identify disallowed content, and involves human verification through a user interface to validate the presence of such content.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated filtering systems are used to identify disallowed content, then productivity increases, but measurement precision deteriorates leading to false positives and negatives
Solution Approach 1:
The system segments the content filtering task into multiple independent components: automated scanning using bad-word/good-word dictionaries for initial identification, logistic regression scoring for probability assessment, and human reviewer verification for final confirmation. This segmentation allows parallel processing while maintaining accuracy through layered validation.
Solution Approach 2:
The system introduces an intermediary human reviewer layer between automated detection and final content removal. The automated system identifies potential disallowed content and scores profiles, but actual removal decisions are made by human reviewers who verify the automated system's findings, thus maintaining high precision while preserving productivity.
2Measurement precision
If comprehensive content scanning is performed on all member profiles, then measurement precision improves, but loss of time increases due to processing large volumes of data
Solution Approach 1:
The system performs preliminary actions by pre-processing and categorizing words into bad-word and good-word dictionaries before actual profile scanning. This preliminary categorization enables rapid lookup and scoring during the scanning phase, reducing processing time while maintaining comprehensive coverage of all profiles.
Solution Approach 2:
The system implements continuous scanning and scoring of member profiles using automated algorithms that operate constantly without interruption. The logistic regression model continuously evaluates profiles, and the system maintains persistent scanning operations, ensuring all content is eventually reviewed without repeated processing delays.
3Measurement precision
If human reviewers verify all identified profiles, then measurement precision improves, but productivity decreases due to manual review requirements
Solution Approach 1:
The system applies partial human review action by having human reviewers verify only those profiles that the automated system identifies as potential matches for disallowed content, rather than reviewing all profiles. The logistic regression scoring system filters and prioritizes profiles for human review, so reviewers focus only on cases requiring human judgment, maintaining precision while preserving productivity.
4Measurement precision
If multiple dictionaries are used for word classification, then measurement precision improves, but device complexity increases due to multiple data structures
Solution Approach 1:
The system uses a universal logistic regression scoring model that works across multiple specialized dictionaries (bad-word dictionary, good-word dictionary, industry-specific dictionaries). This single scoring mechanism handles classification across all dictionaries uniformly, reducing the complexity of managing multiple specialized processing systems while maintaining high categorization precision through the combined power of multiple dictionaries.
Data Source
AI summary
The disclosed embodiments relate to a system for identifying member profiles containing disallowed content in an online social network. During operation, the system scans member profiles looking up words from one or more fields in “bad-word” and/or “good-word” dictionaries. These dictionaries were previously generated by examining member profiles that were judged by a human reviewer to be “good” profiles or “bad” profiles, and determining whether a word belongs in a specific dictionary based on frequencies of occurrence of the word in the good profiles and the bad profiles. Next, the system computes a score for each member profile based at least partially on occurrences of words from the dictionaries in the member profile. Finally, the system identifies a member profile as containing disallowed content based on the score.