Event Detection Using Title Phrase Clustering for Recall
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current event detection methods suffer from low event recall rates due to short keywords and the formation of impure, oversized clusters when using text bodies for clustering, leading to inaccurate event identification.
Innovation Solution
The method involves acquiring texts with a target keyword, extracting phrases that independently describe event information from titles, and clustering these phrases to form events, thereby improving recall rates and cluster purity by avoiding the use of text bodies for clustering.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If burst detection is used with short keywords, then the detection speed is improved, but the event recall rate deteriorates
Solution Approach 1:
The patent segments the text body into multiple sentences and performs clustering at the sentence level rather than using short keywords or full text bodies. This segmentation allows for more granular event detection that captures complete event information while maintaining detection efficiency.
Solution Approach 2:
The patent introduces sentences as an intermediary between short keywords and full text bodies for clustering. Sentences serve as mediators that contain complete event information while being smaller in scale than full text bodies, thus improving both detection speed and event recall rate.
2Loss of information
If text bodies are used for clustering, then the event information completeness is improved, but the cluster purity deteriorates
Solution Approach 1:
The patent segments text bodies into sentences and performs clustering at the sentence level. This segmentation prevents the formation of oversized clusters containing multiple events while ensuring each sentence contains complete event information, thus improving both cluster purity and information completeness.
Solution Approach 2:
The patent changes the dimension of clustering from full text bodies to sentences, creating a new granularity level. This dimensional change allows for better control over cluster size and purity while maintaining event information completeness through sentence-level event extraction.
3Loss of information
If text bodies are used for clustering, then the event information completeness is improved, but the detection complexity increases
Solution Approach 1:
The patent segments text bodies into sentences, reducing the computational complexity of clustering while maintaining event information completeness. Sentence-level clustering requires less computational resources than full text body clustering while still capturing complete event information.
Solution Approach 2:
The patent changes the clustering dimension from full text bodies to sentences, reducing the scale of data processed during clustering. This dimensional reduction decreases detection complexity and computational burden while preserving event information through sentence-level event extraction.
Data Source
AI summary
A method and an apparatus for event detection, a device, and a storage medium. An event is formed by acquiring a plurality of texts including a target keyword; extracting phrases independently describing event information from titles of the plurality of texts; and clustering the extracted phrases and gathering texts where phrases belonging to a same cluster are located to form an event. The accuracy of the event detection and the recall rate for an event can be improved through the method provided by embodiments of the present application.