Topic-driven patent recommendation method and system, equipment and storage medium

Abstract

The invention provides a topic-driven patent recommendation method and system, equipment and a storage medium, and the method comprises the steps: obtaining topic information, inventor information and applicant information of candidate patents and to-be-tested patents, obtaining sampling words according to the topic information, and enabling an inventor, an applicant and the sampling words to be correlated with topics; constructing a topic model, inputting associated inventors, applicants, sampling words and topics into the topic model, training the topic model to form a dynamic topic model, inputting topic information of a to-be-tested patent into the dynamic topic model, obtaining text feature representation of the to-be-tested patent, inputting the text feature representation into the trained word embedding model, and obtaining the text embedding amount of the to-be-tested patent and the candidate patents, calculating the similarity of the candidate patents, and performing sorting according to the similarity to obtain similar patents of the to-be-tested patent. According to the method, the effective range of patent matching can be expanded, and the relevance between words can be improved, so that the patent recommendation accuracy is improved.

Description

technical field [0001] The present invention relates to the technical field of patent recommendation, in particular to a topic-driven patent recommendation method, system, device and storage medium. Background technique [0002] Patent recommendation is a technique for recommending patent documents that contain similar content to a given target patent. In recent years, automatic patent recommendation for the purpose of assisting in patent retrieval, analyzing patent documents, applying for new invention patents, etc. has become a new popular research field. [0003] The current mainstream patent recommendation algorithms mainly include the following categories: [0004] (1) Content-based recommendation. It mainly finds the correlation between patents based on the metadata of recommended patents, and then recommends similar patents to users based on the user's previous preference records. [0005] (2) Recommendation of collaborative filtering. It mainly uses the preferenc...

AI Technical Summary

Problems solved by technology

Attempting to combine convolutional neural networks with word embedding methods to discover patents, the embedding model learns each word or document with one embedding result, which leads to the inability to distinguish multiple meanings of the same word in different contexts

[0013] In summary, the patent recommendation methods in the prior art cannot take advantage of the heterogeneity of patents to enhance the distinction or association between words, resulting in low accuracy of patent recommendation

Images