| The patent document data set is currently the largest set of technical information in the world,which includes the technical achievements and development trends in almost all application fields and plays an important role in the fields of technology,business and law.With the rapid development of economic globalization,science and technology,and the increasing awareness of the internationalization of intellectual property protection,patent conflicts and barriers have deeply troubled domestic enterprises and R&D institutions.The information of patent technology innovation has become the focus of attention,and the demand for tracking and researching foreign patent technology is increasing every day.Therefore,it is necessary to obtain patent information of other countries in a timely and accurate manner.So that cross-language patent recommendation research as an effective way to obtain patents related to other languages has received widespread attention.Crosslanguage patent recommendation can help enterprises and individuals select relevant important patents,discover relevant technology development trends,track the latest technological progress,and then provide personalized information recommendations and decision support services.At present,patent recommendations are mainly based on monolingual patent document data,and similar patent recommendation in monolingual environment is studied from the perspective of technical keywords,topics and other content characteristics of patents.However,there are few cross-language patent recommendation and analysis specifically for bilingual or multilingual patent literature data.Cross-language patent recommendation and analysis generally apply the techniques of word translation and machine translation involved in cross-language retrieval directly to patent recommendation,mainly including three methods: dictionary-based method,corpus-based method and machine translation method.The above-mentioned cross-language patent retrieval and recommendation methods are mostly from the perspective of accurate translation of query words and texts,and often require large-scale bilingual dictionaries,bilingual corpora,and machine translation methods to achieve supervised cross-language query expansion.As a result,it is difficult to extend the application of these methods to other fields for cross-language patent recommendation.At the same time,most of the recommended patents are similar patents,and the diversity and relevance of recommendations need to be further expanded.Relevant patent recommendations need to be studied from the perspective of patent text semantics to provide better decision support services.In this paper,Chinese and English patent word vectors based on the word representation learning was trained,and an unsupervised cross-language word vector mapping method was designed to map them to a unified semantic vector space through linear transformation so as to form a semantic mapping relationship between Chinese and English words.Based on this,the text representation learning method was used to automatically learn the semantic information of Chinese and English patent texts and express them in a unified semantic vector space.Finally,the vector similarity calculation method was used to calculate the semantic similarity between patent texts in different languages,to construct a cross-language patent recommendation method based on representation learning in a bid to achieve recommendation of cross-language relevant patents.Experiments related to "wireless communication" show that this method can achieve comprehensive and accurate Chinese-English cross-language patents recommendation and the accuracy rate of the top 1 and the top 5 recommended patents reached 55.63% and77.82%,respectively,which are improved compared with the weak supervised recommendation method.This method can be expanded to the research and application of patents recommendation in other domains and languages. |
