Research And Practice Of The Text Processing Technology In Electronic Bidding

With the maturity of computer technology,electronic bidding and bidding develops rapidly,and the form of bidding evaluation gradually transforms from offline bidding evaluation to online bidding evaluation by experts.However,the bid evaluation method of "expert evaluation system" still has its own limitations.One is that the content of the bid book is huge,and the experts are limited in energy and physical strength,so they cannot carry out effective comparative analysis on the bid book within a limited time to judge whether there is a bid crossover behavior.Secondly,the manual integration of expert review opinions is time-consuming and laborious,and there are subjective problems,which is not conducive to the final decision of the tenderee.Aiming at the above problems,this paper studies the text processing technology,constructs the bidding text similarity calculation model and the expert opinion integration model,in order to provide the convenience for the experts to review the bidding documents,and completes the expert opinion integration.The bidding text similarity calculation model constructed in this paper is based on deep learning theory.The basic architecture of this model is the twin network.In this paper,the word vector trained by Cw2 vec model is proposed in the input layer,and the part of speech features and dependency features are integrated to enrich the semantic information of bidding text.In the network layer,a three-layer network structure is used,with double-layer Bi LSTM and Selfattention.In order to make the model training effect better,the output results of the first layer Bi LSTM and Self-attention are stitched together as the input of the second layer Bi LSTM.Finally,the cross entropy loss function is used to optimize the cosine similarity distance.Experimental verification shows that Cw2 vec,part-of-speech feature and dependency feature can well represent the words in the bidding text,and double-layer Bi LSTM and self-attention can effectively enhance the feature representation of word vector.Finally,the similarity calculation accuracy and F1 value of this model are good.The expert opinion integration model constructed in this paper is based on machine learning and decision making methods.In the pre-processing stage,according to the grammar and style characteristics of Chinese,eight groups of SAO extraction rules are put forward to extract the subject-action-object structure of opinions.In the section of opinion labeling,a two-step fusion opinion classification method is proposed to label topics and emotional tendencies.Finally,the non-explanatory opinions are fused based on the fuzzy comprehensive evaluation method,and the explanatory opinions are fused based on the sentence fusion method to get the final integrated opinions.The experimental results show that the extraction rules are more effective,and the opinion classification method based on two-step fusion has a good accuracy.The model in this paper can finally effectively complete the integration of expert opinions and reduce the disadvantages of manual integration.