Joint Sentiment-Topic Modeling Without Lexicons
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing NLP systems are inadequate for performing joint sentiment-topic (JST) modeling, failing to efficiently and reliably detect topics, sentiments, and their interactions within individual and multiple documents, and require extensive external lexicon information or prior sentiment labeling.
Innovation Solution
The proposed solution involves using per-document topic distributions, per-document topic-sentiment distributions, and cross-document per-word topic-sentiment distributions to generate topic and sentiment designations for selected words, enabling effective JST modeling without prior sentiment labeling or lexicon information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing NLP systems are used for JST modeling, then basic topic detection may be achieved, but the systems fail to efficiently and reliably detect topics, sentiments, and their interactions within individual and multiple documents
Solution Approach 1:
The patent combines topic modeling and sentiment analysis into a unified joint sentiment-topic modeling framework. The system simultaneously learns topic distributions and sentiment distributions for documents, enabling efficient detection of both topics and sentiments along with their interactions in a single integrated process rather than separate sequential analyses.
Solution Approach 2:
The patent segments the JST modeling process into distinct computational components: per-document topic distributions, per-document topic-sentiment distributions, and cross-document per-word topic-sentiment distributions. This segmentation allows the system to process and analyze different aspects of documents separately while maintaining their interconnections through the unified model.
2Measurement precision
If external lexicon information or prior sentiment labeling is used, then sentiment detection accuracy may improve, but the system requires extensive external resources and prior labeling
Solution Approach 1:
The patent implements a self-service approach where the JST modeling system automatically learns sentiment information and topic structures directly from the document corpus without requiring external lexicon resources or pre-labeled sentiment data. The model self-adapts to the specific domain and vocabulary of the input documents, eliminating dependency on external knowledge bases.
Solution Approach 2:
The patent performs preliminary learning of topic and sentiment distributions during the model training phase, establishing the foundational structures needed for accurate sentiment detection. By pre-learning the relationships between words, topics, and sentiments from the corpus during training, the system prepares the necessary knowledge structures before actual sentiment analysis is performed.
Data Source
AI summary
There is a need for more effective and efficient natural language processing (NLP) solutions. This need can be addressed by, for example, solutions for performing NLP analysis by utilizing joint topic-sentiment (JST) modeling. In one example, a method comprises receiving a per-document topic distribution for a digital document, wherein the per-document topic distribution comprises a per-document topic correlation indication for each candidate topic designation; receiving a per-document topic-sentiment distribution for the digital document, wherein the per-document topic-sentiment distribution comprises a per-document topic-sentiment correlation indication for each topic-sentiment pair of a candidate topic designation and a candidate sentiment designation; generating, based at least in part on the per-document topic distribution and the per-document topic-sentiment distribution, a topic designation and a sentiment designation for each selected word in the digital document; and generating a JST modeling output based a on each topic designation and each sentiment designation.