| Since there is still a gap between the domestic marine engine technology and products with foreign countries.In order to catch up with the international firstclass marine engine,the efficiency of domestic engine research and development needs to be urgent improved.During the design of marine engine products,engineers need to quickly obtain the information which is related to design tasks.However,in the past,the method of patent knowledge reusing was a simple active search,which caused the design process to be cumbersome and inefficient.In order to solve the low efficiency of knowledge reuse in the design of marine engines,this paper conducts relevant research on marine engine patents’ topic mining and patent recommendation algorithms,which can push relevant patents that meet the design requirements of design tasks,and assist engineer in innovative work.This paper mainly researches the following aspects:First,there is a problem of semantic isolation in the traditional patent topic mining.This research plans to measures the similarity of the terms in the patents,then clusters the words with similar technical concepts into one category and bind the terms in the original patent text with corresponding semantic relationship during the topic mining to alleviate semantic isolation and eliminate the problem of uneven contribution of topic information caused by similar semantic words with different frequency levels,and solve the problem that words have important contributions to the topic are ignored due to the low-frequency.Through the improved topic model,high-quality topic features of patents can be obtained,then patents can be represented more accurately to serve subsequent tasks.Secondly,in order to achieve more accurate and effective patent recommendation,the research in this paper also needs to judge the applicability of patents and design tasks to determine the range of patents that may meet the task requirements which can use for improving the accuracy of the patent recommendation results.In this part of the research,first of all,in addition to using the natural characteristics of patent itself such like the topic characteristics,it is also necessary to supplement the unnatural characteristics of the patent to enrich the patent features which are used for applicability analysis.Then,by analyzing the characteristics of the patents which are applicable to the current design task,the multi-level and deep relationship between its natural features and unnatural features is conducted to obtain the relationship that how can they influent the applicability of patents and design tasks,so as to analyze the unread patents and determine that whether them are applicable to the design task or not and determine the range of patents that can be recommended.Thirdly,in order to achieve accurate recommendation of patents and avoid the problem that engineer still need to perform manual screening,it is also necessary to sort the final recommended patents by the relativity with current task.The traditional collaborative filtering algorithm has the problem of insufficient use of rate information,resulting in poor recommendation performance.This research plans to amend the collaborative filtering algorithm by introducing a variety of patent measurement methods.First,it analyzes the patent rate data in each design task,then obtains the vector representation of each patent by embedding patents to dense vectors,then the similarity factor of the patent can be obtained by analyzing corresponding vectors.Then the specific rate data is introduced into the calculation of patent relativity with analyzing the positive and negative evaluation of the patent,then we can get the rate penalty factor subsequently.With combining the above two correction factors,we can get an advanced relativity calculation method.During a specific task,we can calculate and rank the relevance of unread patents with rated patents within the scope of patents that satisfy the applicability,then recommend high ranking patents finally.Finally,this research also needs to develop the prototype system of the marine engine patent recommendation system software,this part of research will be conducted by analyzing the functions required in actual engineering,conducting the algorithm deployment of this research content and writing the software interface which is convenient for the engineer to operate,and finally complete the research of this paper. |
