Sentiment Analysis Normalization for Social Media Signal Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current social media analytics techniques are limited by their reliance on coarse-grained sentiment estimates and activity metrics, which are sensitive to thresholds and unsuitable for time series normalization, making it difficult to compare entity sentiment levels and detect significant changes or deviations in social media activity across different entities.
Innovation Solution
A system that measures and normalizes sentiment and activity levels on a continuous scale, using a vector of sentiment metrics including sentiment time series scores, smoothed averages, volume changes, and diversity of posting accounts, enabling entity-to-entity comparisons and detection of significant changes relative to a sentiment universe or market index.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If coarse-grained sentiment estimates using Natural Language Processing are employed, then sentiment analysis can be performed, but the estimates are highly sensitive to thresholds and not suitable for time series normalization
Solution Approach 1:
The patent transforms discrete sentiment categories (positive, negative, neutral) into continuous numerical values with multiple decimal places. This parameter change enables precise measurement while maintaining ease of analysis, resolving the contradiction between computational simplicity and measurement precision.
Solution Approach 2:
The invention adds a temporal dimension by creating time series of sentiment values and applying normalization techniques. This transforms static coarse-grained estimates into dynamic normalized metrics that can be compared across time and entities, solving the suitability issue for time series analysis.
2Quantity of substance
If activity metrics are calculated by counting total mentions, then the volume of social media activity is measured, but significant deviations from normal levels cannot be detected and comparison between entities is not useful
Solution Approach 1:
The system performs preliminary actions by calculating historical averages and standard deviations of mention volumes before detecting current activity levels. This preprocessing enables the detection of significant deviations by comparing current metrics against established baselines, solving the problem of detecting unusual activity patterns.
Solution Approach 2:
The patent transforms raw mention counts into normalized activity metrics by applying statistical transformations (mean, standard deviation, z-scores). This parameter change enables meaningful comparison between entities with different baseline activity levels and detects significant deviations from normal patterns.
Data Source
AI summary
A system for scoring micro-blogging messages is provided, including an extractor, and evaluator, a calculator, and a publisher. The extractor may be configured to receive micro-blogging messages, to detect messages containing terms of interest, to extract raw data, and to store the data in a database. The evaluator may be configured to access and parse the stored data into tokenized data, and to store the tokenized data in a database. The evaluator may also be configured to identify relevant micro-blogging messages; to tag message as indicative; and to filter messages from low-volume or malicious sources before being tagged as indicative. The calculator may be configured to access a sentiment dictionary; to calculate a sentiment score of the tokenized data, and to calculate a sentiment signature for a term of interest. The publisher may be configured to provide access to clients of the system.