Clustering method for unsupervised learning of Chinese comments, computer program product and server system

Abstract

The invention provides a clustering method for unsupervised learning of Chinese comments, a computer program product and a server system, and the clustering method comprises the steps: obtaining comment data, and sorting to obtain a corpus; preprocessing the comment content information in the corpus, and carrying out word segmentation and word vector training; extracting candidate tags; performingduplicate removal processing on the candidate tag library; carrying out sentiment word filtering on the candidate tags subjected to duplicate removal; carrying out DBSCAN-based clustering operation on the candidate tags without the invalid tags to obtain the magnitude of all the candidate tags, and carrying out descending order arrangement on clustering results according to the number; and finally, counting each clustering magnitude, and outputting TopN. According to the method, the clustering mode based on unsupervised learning is provided, the problem that a comment result is difficult to objectively express through an existing label clustering method is solved, autonomous and unsupervised extraction and learning can be carried out according to comments and the actual content of labels,and the clustering result which reflects the real comment result more objectively is provided.

Description

technical field [0001] The invention relates to the technical field of data mining and processing, in particular to a clustering method for unsupervised learning of Chinese comments, a computer program product and a server system. Background technique [0002] At present, in the evaluation of goods or services on e-commerce platforms or forums, tags are often extracted and displayed through technical means, so that potential users can directly obtain the most direct evaluation of products or services. There are two main ways to generate these tags. One of them is extraction, which is to extract the words or phrases with the highest frequency based on statistical principles, form tags, and arrange them in order of frequency. This method is used when tagging It will generate a lot of noise, and the extraction based only on statistical principles often gets strange results (labels), which cannot truly reflect the characteristics of reviews or products; the other is based on the...

AI Technical Summary

Problems solved by technology

There are two main ways to generate these tags. One of them is extraction, which is to extract the words or phrases with the highest frequency based on statistical principles, form tags, and arrange them in order of frequency. This method is used when tagging It will generate a lot of noise, and the extraction based only on statistical principles often gets strange results (labels), which cannot truly reflect the characteristics of reviews or products; the other is based on the generation of pre-customized labels, and then comment information If it occurs once, it will be accumulated by 1. After querying all the comments, you will get the cumulative result of the custom label. Take the first N and arrange it to get the final labeling result. This method often requires comparative labor when labeling. Low efficiency, and can only be accumulated for custom tags, often ineffective for new comments or keywords

[0003] Combining the above two methods, both are based on supervised clustering, which is characterized by the fact that it is difficult to reflect the real situation

Images