Research On Online Short-rent Trust Feature Selection And Computing Based On Multi-modal Data

In recent years,the sharing economy has risen and developed rapidly.Online short-rent is one of the representative industries of sharing economy and has received great attention from academic,commercial and other circles.Different from traditional e-commerce transactions,online short-rent transactions are completed based on cooperation concept by unfamiliar host and guest with guidance by third-party platforms.Trust between host and guest plays a core role in such transactions.Plus,China’s credit system is under construction,researches on online short-rent trust are of great significance to guard the trading environment and maintain the healthy and sustainable development of economy.However,there are few researches on trust feature selection and computing in online short-rent for now.Improvements could be made in integrating multi-modal data,obtaining core trust features,and the comprehensiveness of trust computing.From the perspective of online short-rent platforms,this study performes trust feature selection and trust computing based on multi-modal data,obtaines core trust features through feature selection,evaluates and predicts hosts’ perceived trust.Thus,helps to improve the online short-rent trust mechanism as well as to maintain a friendly environment and to improve trading experiences,boost the prosperity and development of online short-rent economy.This study performs trust feature selection and computing based on online short-rent multi-modal data.Including:1.Trust feature extraction integrating image,text and numerical multi-modal data.In online short-rent transactions,guests not only pay attention to numerical data like price,but also pay more attention to pictures and texts which present the image and characteristic of hosts.Therefore,this study integrated three kinds of online short-rent multi-modal data to extract trust features,especially image and text based on "visual trust" and other theories.Thus,trust features could be enriched and description angels of features could be expanded.Subsequent experiment results show that image and text features account for about 50% of the core trust features,showing that they have great impacts on host’s perceived trust.2.Core trust feature recognition based on efficient feature selection.Identifying core trust features from high-dimensional features considering problems caused by missing values in data are very important in online short-rent trust analysis.Therefore,this study designed two-stage trust feature selection based on tolerance rough set theory.In the first stage,an efficient algorithm was designed to obtain low-dimensional feature subsets,which can solve the problem of low efficiency in analyzing massive high-dimensional data with missing values.In the second stage,trust features were further screened to ensure their effects on perceived trust.Experiment results show that the designed algorithm can effectively reduce the dimension and obtain above 95% of accuracy using about 10% of all features.Compared with the existing algorithms,the computing efficiency is improved by more than 90%.3.Construction of online short-rent trust computing framework facing the dynamic changes of core trust features.Trust computing should consider the dynamic changes of core trust features with time and the handling of missing values.Therefore,this study constructs an online short-rent trust computional framework.By updating trust feature subsets through feature selection,the dynamic changes of core trust features can be tracked.By providing multiple low-dimensional trust feature subsets to evaluate and predict host’s perceived trust,the problem of model failure caused by missing values could be alleviated.Experiment results show that comparied with the trust computing model of an existing research,the trust evaluating performances of ours were improved by about 8%,up to 13%.Predicting performances were improved by 5%,considering accuracy,precision,etc..At last,research works and limitations were summarized with advices given for future researches on online short-rent trust.