A Query Expansion Research Based On Word Vectors About Agricultural Knowledge

Vertical search is a special search method for professional field.In terms of agriculture,amount of technology and experience can be provided to users through agricultural vertical search engines.Although agricultural vertical search engines have been able to solve the problem of specialized retrieval of agricultural information,due to the fact that all parts of China have their own dialects,and that agricultural producers often have typo due to unskilled typing,when using search engines,It is possible to use dialect vocabulary or typo,which have not been taken seriously and resolved in current agricultural vertical search engines.The purpose of this paper is to use word vector and fixed-point matching to implement a query expansion to solve the problem of low recall rate when dialect vocabulary or typo occurs while agricultural producers use search engines.When there is a dialect vocabulary or a typo word in the query,search engine recall rate and precision rate are very low,and it is very likely that no relevant documents can be found,which will greatly impair the user's enthusiasm for using search engines..Therefore,in view of this situation,it is necessary to increase the recall rate and precision rate of the search engine.Traditional thesaurus methods cannot exhaust the error situation and cannot completely solve these abnormal query statements.The word clustering method cannot find its class because abnormal words often do not exist in the corpus.Pseudo-related feedback methods cannot be resolved because the initial query document is likely to be completely irrelevant.Relevant feedback needs the cooperation of users.However,the majority of the people who are still engaged in farming in China are middle-aged and elderly people,and they are impatient with new things,so it is not reasonable.The query log method cannot be resolved for the initial cold start phase.The fixed-point matching query expansion method based on the word vector proposed in this paper firstly diagnoses the abnormal words in the query sentence,and finds the abnormal words by comparing the user's query statement with the degree of freedom and the constraint degree.Afterwards,it is judged whether the abnormal word is keyword or not,and the difference between the title document set and the content document set of keyword is mainly used to judge.The word vectors and local planting information generated by the corpus training language model are reused to return the most likely extended word of the abnormal word according to the normal vocabulary in the query sentence,thereby increasing the recall rate of the search engine.In the experimental stage,a data set was compared with the traditional two methods,and the recall rate and precision rate in the existence of dialect vocabulary and typos were tested.Experiments show that the query expansion method of this paper can improve the search engines performance when abnormal words occur.