Research On Structured Extraction Of Recruitment Text Data Based On Deep Learning

The employment problem has always been an economic and social problem concerned by the society and the country,and in solving the employment problem,the acquisition of employment information is a very important work.In recent years,the employment and recruitment information on the Internet has exploded.In the face of massive recruitment information,it is very important to accurately screen out the key information.Structured extraction of recruitment information can help to obtain the key information in the recruitment information,which has practical significance for the stable development of the market.Information extraction aims to extract key information from huge amounts of data,and stored in a structured form.Named entity recognition is an important information extract subtasks.The methods of named entity recognition,from rule-based and dictionary based to statistical probability based,require a lot of manpower for dictionary construction or feature statistics,and need a lot of labor and time cost.However,the method based on deep learning does not need too many artificial features,and can autonomously learn text information,which has made some progress in the task of named entity recognition.In this paper,the named entity recognition method based on deep learning is used to identify the named entity of the recruitment text information released by colleges and universities.The named entity objects are mainly the name of the recruitment position and the name of the recruitment major.Specific work is as follows:Firstly,the crawler tool is used to collect the recruitment information published by the university employment information network,and the information is preprocessed.In the data preprocessing process,in order to strengthen the effect of Chinese word segmentation,the recruitment position NOUN database and recruitment position NOUN database are constructed.Secondly,the lstm-crf model is constructed.The model mainly uses LSTM and CRF to learn semantic features and adjacent label relationship respectively,and the model is used to recognize the named entity of the collected data.Through the simulation experiment,the F1 values of entity recognition of post name and professional name reach 82.52% and 89.23%respectively.Thirdly,to solve the problem of context information and semantic dilution,the BiLSTM-Att-CRF model is constructed by using bidirectional LSTM to replace LSTM network and adding attention mechanism.In this model,bidirectional LSTM uses forward and backward LSTM to learn context to acquire bidirectional semantic dependence,and uses attention mechanism to focus the model on local features to solve the problem of semantic dilution.In this paper,the data collected and processed are used to train and test the proposed model.The experimental results show that compared with the benchmark model,the F1 value of the improved model in entity recognition of post name and professional name is increased by 5.53% and 5.31% respectively.