Evaluation Of Text Mining Algorithms Based On Multi-attribute Decision-making And Its Application In Product Ranking

The development of the Internet has resulted in a substantial increase in usergenerated content,including text data.These text data contain a lot of valuable information.Although people can understand the text data,the amount of text data exceeds the upper limit that human can handle.In order to make full use of this information,people need to resort to methods that can automatically mine this information.Text mining is precisely this method,and many text mining researches have been proposed,which makes the text mining technology develop rapidly.However,in some more complex scenarios,there are still problems that are difficult to solve with existing research.In this paper,two complex decision problems in text mining are solved from the perspective of algorithm and application by combining multi-attribute decision method.These two problems include:The first is the key algorithm selection problem in text classification.The variety of text data and the large volume of capacity make a single algorithm evaluation system and simple application unable to meet the demand.Text classification is also facing the same situation.Text classification is one of the most important methods in text mining.After a long period of development,many methods have been proposed.When using text classification to solve a special problem,you first need to choose the appropriate text classification method.This requires evaluation of text classification methods,and the evaluation of these methods usually involves more than one evaluation indicator,which reflects different aspects of algorithm performance.It is difficult for an algorithm to perform optimally on all evaluation indicators.Only using any one evaluation indicator to completely measure the quality of an algorithm is biased,which makes the algorithm selection in text classification a complicated decision-making problem.Especially in the text classification problem of small samples,due to the high-dimensional characteristics of text classification,in addition to the performance of the algorithm,the stability is also an important measurement standard.The second is the product ranking problem based on text mining.Through product ranking,companies can determine product status,determine pricing and competitive strategies;consumers can use ranking as a reference to guide shopping decisions.Although a small number of scholars have proposed product ranking methods based on text mining methods and online reviews,they can achieve product rankings with less human participation,and this method is often more scalable than traditional methods.Their ideas are as follows: first,text mining based on online reviews gets consumers' sentiment direction for product attributes,then generates product attribute scores based on sentiment directions,and finally uses multiattribute decision making to comprehensively score products to obtain product rankings.However,the existing method does not consider the content of reviews on multiple platforms,and the product ranking based on the analysis of a single platform can only represent the consumer opinions of one platform.The product rankings generated by product reviews from different platforms are different,and it is biased to use a single platform to reflect the market position of the product.In order to solve the above problems,we introduce the multi-attribute decisionmaking method and its fuzzy group decision-making form as a solution,respectively solving the multi-attribute decision-making problem in algorithm evaluation and the multi-platform integration problem in product ranking.For the first problem,this paper tests the effect of different multi-attribute decision-making methods in evaluating key algorithms for text classification,and then analyzes the regularity of these key algorithms in small sample text classification problems.For the second problem,we propose a cross-platform product ranking method,which can integrate the review content of multiple platforms using multi-attribute group decision-making methods and then generate product rankings based on fuzzy multi-attribute decision-making methods.The specific research content and related conclusions are as follows:(1)First,we use the multi-attribute decision method to solve the evaluation problem of key algorithms for text classification.Specifically,it includes the following two problems: first,the evaluation of feature selection methods in small sample text classification involves multiple indicators;second,in the traditional text classification process,it is necessary to simultaneously select feature selection methods,the number of features,and the classifier.Experimental results show that the multi-attribute decisionmaking method can effectively solve these two problems.In addition,according to the experimental results,we found some rules of feature selection methods and classifier performance,which can provide a certain reference for future generations to face similar problems.(2)Aiming at the problem of product ranking research based on text mining without considering multiple platforms,a cross-platform product ranking method based on multi-attribute decision-making is proposed.This method can integrate the review content of different platforms and convert the review content into usable.Compared with a single platform,the decision matrix for evaluating the market position of products provides product ranking results that better reflect the views of the entire market.In addition,based on our proposed ranking method,we propose a priority generation method for product attribute improvement.The experimental results prove the effectiveness of the cross-platform product ranking method and product attribute improvement priority generation method proposed in this paper.Based on our experimental analysis on the mobile phone market,we have made product improvement and marketing recommendations for different mobile phone brands.(3)In response to the problem that the product ranking method we proposed can only integrate the reviews of platforms with complete product and attribute information,we have extended the method in(2)to a certain extent,and can effectively deal with the decision matrix with missing values.The integration of different platforms in the case of missing values is more general than the method proposed in(2),and it also has a certain reference value for other group decisionmaking problems with missing values.By introducing platform reviews with missing values,our experimental results show that our method can effectively integrate the content of these platforms with missing reviews.(4)In view of the difference in the market position ranking of products on different platforms,we analyzed the reasons for this difference and then put forward suggestions for platforms and merchants to improve the market position of products.Among them,the reason is mainly analyzed from two aspects,the format of the comment and the platform operation mode.We judge whether the review format has an impact on the ranking of the market status by comparing whether there is a difference in the ranking of product market positions generated by retaining and removing comments in different formats.The results show that there is no obvious change in the ordering of the format operation,which excludes the influence of the comment format.After that,based on the discussion of the differences in the operating modes of different platforms,we believe that the operating mode may be the reason for the difference in product rankings.In response to this,we have put forward a certain degree of advice to both the platform and the merchants.In the past,people rarely used multi-attribute decision-making methods to solve complex decision-making problems in text classification.Our research made up for this shortcoming.In terms of method evaluation,we have found some rules for the performance of text classification methods,which can give other researchers a certain degree of reference on similar issues;in terms of product ranking,we propose a method that rarely reviews the content of multiple platforms.The articles considered together have some groundbreaking significance.At the same time,the method we proposed also has certain reference value for understanding consumer preferences.