Comment usefulness prediction method based on comment picture aesthetics

By constructing a nonlinear regression model, analyzing the relationship between the aesthetics of comments and the usefulness of comments, combined with the depth of comment text, the problem of the neglected aesthetics of comments in the existing technology is solved, and efficient prediction of comments and online predictions are achieved. Reputation management.

CN119006070AInactive Publication Date: 2024-11-22HARBIN INST OF TECH
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411114230.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-14
Publication Date
2024-11-22
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The prior art ignores the important role of comment photos in online comments, resulting in information overload and low efficiency in predicting comments’ usefulness, and the inability to effectively utilize the impact of comment image aesthetics on usefulness.

Method used

By constructing a nonlinear regression model based on TripAdvisor hotel review data, we analyze the relationship between the aesthetics of the comment image and the usefulness of the comment, and conduct differential research based on the depth of the comment text, we provide a prediction method for comment usefulness based on comment image aesthetics.

Benefits of technology

This has realized an inverted U-shaped nonlinear impact analysis of the aesthetic level of comment pictures and the usefulness of comments, providing practical guidance to help online comment platforms and hotel managers improve online reputation management efficiency, reduce costs, and improve corporate performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119006070A_ABST
    Figure CN119006070A_ABST
Patent Text Reader

Abstract

The invention discloses a comment usefulness prediction method based on comment picture aesthetics, and belongs to the technical field of data analysis. The method comprises the steps of 1, obtaining online comment data of a target hotel; 2, screening and cleaning the data to obtain a data set; 3, determining a research model of the comment usefulness influence of the comment picture aesthetics, and defining and calculating variables required by the model; 4, setting an explained variable, an explaining variable, an adjusting variable and a control variable, constructing a nonlinear regression model, and calculating to obtain a coefficient and a residual term of the nonlinear regression model; and 5, replacing a least square regression method with a Tobit regression method, selecting a sub-sample set to carry out robustness test, and obtaining a comment usefulness prediction model based on comment picture aesthetics according to the coefficient and the residual term of the nonlinear regression model calculated in the step 4. According to the method, comprehensive, efficient, high-accuracy and easy-to-implement mining and analysis can be carried out on the user comments and the hotel information.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of data analysis, and in particular relates to a method for predicting the usefulness of comments based on the aesthetics of commented pictures. Background Art

[0002] Online reviews can help consumers reduce the uncertainty and risks associated with online shopping and become one of the most trusted sources of information in consumers' daily online shopping experience. However, with the rapid development of social media, the surge in online reviews has led to information overload and hindered effective information processing. In order to help consumers find useful reviews effectively, many online review platforms have adopted a usefulness voting system. Review usefulness reflects the overall quality and information content of online reviews, which helps to improve the trust of online reviews and promote information processing, especially when the information load is high. Therefore, it is crucial to explore the prediction method of review usefulness.

[0003] Photos and text are both important components of online reviews, and together they influence potential consumers' views on the usefulness of reviews. As a supplement to text reviews, photos provide visual clues for tourist destinations or service agencies, thereby attracting consumers' attention. Photos can convey richer information than text content, and are more vivid, more eye-catching, and easier to understand. However, most related inventions focus on the prediction of the usefulness of online reviews by text content and reviewer-related features, while ignoring the important role of review photos. Summary of the invention

[0004] The purpose of the present invention is to solve the above problems existing in the prior art and to provide a method for predicting the usefulness of comments based on the aesthetics of comment pictures.

[0005] The present invention uses user review data of hotels on TripAdvisor.com, an international mainstream online travel platform, to perform data analysis. The present invention constructs key variables such as the aesthetics of review images, the depth of review text, and the usefulness of reviews, and further constructs a nonlinear regression model to explore the impact of the aesthetics of review images on the usefulness of reviews, as well as the differences at different levels of review text depth. The data results verify the inverted U-shaped nonlinear impact of the aesthetics of review images on the usefulness of reviews, and calculate the specific inflection point. At the same time, the impact is more significant when the depth of the review text is low. The above invention provides useful guidance for identifying the relationship between the aesthetics of review images and the usefulness of reviews, which is beneficial for platforms and hotel managers to conduct online reputation management.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] The method for predicting the usefulness of comments based on the aesthetics of the commented pictures comprises the following steps:

[0008] Step 1: Obtain the original data set; obtain the online review data of the target hotel to obtain the original data set;

[0009] Step 2: Data preprocessing: Perform data preprocessing on the original data set. First, filter out the review data that contains at least one review image, then delete the samples with missing data and abnormal data to obtain the preprocessed data set.

[0010] Step 3: Variable definition and calculation: sort out and summarize the relevant literature in the field of online reviews, determine the research model of the impact of review usefulness on review image aesthetics, define the variables required for the research model, and use the preprocessed data set obtained in step 2 to calculate the variables required for the research model;

[0011] Step 4: Nonlinear regression model analysis: Based on the research model, the defined variables are set as explained variables, explanatory variables, moderating variables and control variables, and a nonlinear regression model containing the above variables is constructed. Then, the variable data obtained by calculation in step 3 is imported into STATA software to fit the model, and the coefficients and residual terms of the nonlinear regression model are obtained.

[0012] Step 5: Robustness test; the nonlinear regression model in step 4 is tested for robustness by replacing the least squares regression method with the Tobit regression method and selecting a subsample set for testing; if the nonlinear regression model passes the robustness test, it means that the model is reliable, and then according to the coefficients and residual terms of the nonlinear regression model calculated in step 4, a prediction model of the usefulness of comments based on the aesthetics of the commented pictures is obtained; if it fails the robustness test, the nonlinear regression model needs to be rebuilt and analyzed, that is, return to execute step 4.

[0013] Furthermore, in step 1, online review data of the target hotel is obtained from the TripAdvisor website through a crawler program.

[0014] Furthermore, the crawler program is used to obtain online review data of the target hotel from the TripAdvisor website, including usefulness votes, review scores, review text, review pictures, review time, check-in time, whether management feedback has been received, travel type, user experience and hotel ID.

[0015] Furthermore, in step three, the variables required for the research model are defined, including review helpfulness, review image aesthetics, review text depth, review sentiment, rating, text readability, number of review images, management feedback, time interval, user experience, travel type, year-month fixed effects of the time when the review was posted, and hotel fixed effects.

[0016] Furthermore, the variables of review usefulness, review text depth, rating, number of review images, management feedback, time interval, user experience, travel type, year-month fixed effects of review posting time, and hotel fixed effects are obtained from the TripAdvisor website, and the three variables of review image aesthetics, review sentiment, and text readability need to be calculated with the help of tools; the calculation process is:

[0017] First, the aesthetics of the commented pictures variable is calculated using a multi-scale image quality converter to calculate the aesthetics score of each picture. Then, the average aesthetics score of all pictures in each comment is used as a measure of the aesthetics of the commented pictures.

[0018] Secondly, the comment sentiment is measured by the following formula (1):

[0019]

[0020] Among them: i represents hotel, j represents comment, Emotion ij represents the sentiment score of the jth review of the i-th hotel, PW ij NW represents the number of positive sentiment words in the jth review of the i-th hotel. ij TW represents the number of negative sentiment words in the jth review of the i-th hotel. ij represents the total number of words in the jth review of the i-th hotel; the above sentiment word frequencies are obtained with the help of LIWC software, and the initial sentiment value scores of each review calculated according to formula (1) are between -1 and 1;

[0021] Finally, the text readability was calculated using the Gunning FOG index in Python software. The larger the Gunning FOG index, the worse the text readability.

[0022] Furthermore, in step 4, the defined variables are set as explained variables, explanatory variables, moderating variables and control variables, specifically:

[0023] The helpfulness of reviews was taken as the explained variable, the aesthetics of review pictures as the explanatory variable, the depth of review text as the moderating variable, and review sentiment, rating, text readability, number of review pictures, management feedback, time interval, user experience, travel type, year-month fixed effects of time, and hotel fixed effects as control variables.

[0024] Furthermore, in step 4, the nonlinear regression model construction process is: the explained variable is expressed as a function of the explanatory variable, the quadratic term of the explanatory variable, the constant term, the error term, the hotel fixed effect and the time fixed effect; the expression of the nonlinear regression model without a moderating variable is shown in the following formula (2):

[0025]

[0026] Where: i represents the hotel, j represents the review, Lg Helpfulness ij represents the logarithm of the number of helpfulness votes obtained by the jth review of the i-th hotel, Aesthetics ij Aesthetics represents the average aesthetic score of the pictures included in the jth review of the i-th hotel. 2 ij represents the quadratic term of the average aesthetic score of the pictures included in the jth review of the i-th hotel, Control ij represents the control variable for the jth review of the i-th hotel, λ ij The coefficient of the control variable for the jth review of the i-th hotel, Hotel i represents the fixed effect of the i-th hotel, y represents the year of the review, m represents the month of the review, and Time ym represents the year-month fixed effect of the time of posting comments, ε1 is a random disturbance term, β 10 represents the constant term, β 11 represents the linear influence of the aesthetics of the review image on the usefulness of the review, β 12 Indicates the nonlinear influence of the aesthetics of the review image on the usefulness of the review; focus on β 11 and β 12 , if β 11 is significantly positive, and β 12 If it is significantly negative, it means that the aesthetics of the review image has an inverted U-shaped effect on the usefulness of the review.

[0027] Furthermore, the review text depth is introduced into the model as a moderating variable. At this time, the explained variable is expressed as a function of the explanatory variable, the quadratic term of the explanatory variable, the moderating variable, the interaction term between the explanatory variable and the moderating variable, the interaction term between the quadratic term of the explanatory variable and the moderating variable, the constant term, the error term, the hotel fixed effect and the time fixed effect. The expression of the nonlinear regression model with the moderating variable is shown in the following formula (3):

[0028]

[0029] Where: i represents the hotel, j represents the review, Lg Helpfulness ij represents the logarithm of the number of helpfulness votes obtained by the jth review of the i-th hotel, Aesthetics ij Aesthetics represents the average aesthetic score of the pictures included in the jth review of the i-th hotel. 2 ijThe quadratic term representing the average aesthetic score of the pictures included in the jth review of the i-th hotel, LgRevDepth ij represents the logarithm of the text depth of the jth review of the i-th hotel, Aesthetics ij ×LgRevDepth ij Aesthetics represents the intersection of the aesthetic score of the jth review image of the i-th hotel and the logarithm of the text depth. 2 ij ×LgRevDepth ij represents the intersection of the quadratic term of the aesthetic score of the jth review image of the i-th hotel and the logarithm of the text depth, Control ij represents the control variable for the jth review of the i-th hotel, λ ij The coefficient of the control variable for the jth review of the i-th hotel, Hotel i represents the fixed effect of the i-th hotel, y represents the year of the review, m represents the month of the review, and Time ym represents the year-month fixed effect of the time of posting comments, ε2 represents the random disturbance term, β 20 represents the constant term, β 21 represents the linear influence of the aesthetics of the review image on the usefulness of the review, β 22 represents the nonlinear influence of the aesthetics of the review image on the usefulness of the review, β 23 represents the linear influence of review text depth on review usefulness, β 24 Indicates the influence of the intersection of review text depth and review image aesthetics on review usefulness, β 25 Indicates the influence of the intersection of the quadratic term of the review text depth and the review image aesthetics on the usefulness of the review; focus on β 25 , if β 25 If it is significant, it means that the depth of comment text has a moderating effect.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] The beneficial effects of the present invention relative to the prior art are as follows: first, the present invention uses the review data from the online travel platform TripAdvisor to construct a nonlinear regression model, analyzes the nonlinear relationship between the aesthetics of the review image and the usefulness of the review, and the influence of the interaction between the aesthetics of the review image and the depth of the review text on the usefulness of the review, and obtains a prediction method for the usefulness of the review using the aesthetics of the review image based on the model analysis results. This provides practical suggestions for online review platforms and service industry managers, helping them to improve the perceived value of online reviews by potential consumers at a lower cost, thereby improving corporate performance. Secondly, the prediction model proposed in the present invention can calculate the optimal level of the aesthetic effect, that is, the inflection point of the inverted U-shaped function. The online review platform can construct an online review usefulness prediction model for a specific hotel based on the present invention, and calculate the optimal level of the aesthetic effect based on the data of the specific hotel and the unique characteristics of each hotel. Finally, the prediction method proposed in the present invention can also analyze a large amount of hotel review data, and conduct comprehensive, efficient, high-accuracy and easy-to-implement mining and analysis of user reviews and hotel information. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a flowchart of the method for predicting the usefulness of comments based on the aesthetics of comment pictures of the present invention. DETAILED DESCRIPTION

[0033] The technical solution of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0034] Specific implementation method 1: Figure 1 As shown, this embodiment records a method for predicting the usefulness of comments based on the aesthetics of comment pictures, and the method includes the following steps:

[0035] Step 1: Obtain the original data set; obtain the online review data of the target hotel (such as hotels in the New York area of ​​the United States) from the TripAdvisor website through a crawler program, including usefulness votes, review scores, review text, review pictures, review time, check-in time, whether management feedback was received, travel type, user experience and hotel ID, and obtain the original data set;

[0036] Step 2: Data preprocessing: Data preprocessing is performed on the original data set. First, the review data containing at least one review image is screened out, and then the samples with missing data and abnormal data are deleted to obtain the preprocessed data set (i.e., data screening and data cleaning).

[0037] Step 3: Variable definition and calculation; sort out and summarize the relevant literature in the field of online reviews, determine the research model (nonlinear regression model) for the impact of review usefulness on review image aesthetics, and define the variables required for the research model, including: review usefulness (the number of usefulness votes each review receives), review image aesthetics (the mean of the aesthetic scores of the review images contained in each review), review text depth (the text length of each review, i.e. the number of words contained), review sentiment (the difference in the proportion of positive and negative emotion-related words expressed in each review), score (the total score of each review), text readability (the Gunning value of each review text), and the difference in the proportion of positive and negative emotion-related words expressed in each review. FOG index), number of review images (the total number of images contained in each review), management feedback (whether the review received management feedback), time interval (the interval between each review's check-in and the review's publication, in days), user experience (the experience value of the user account corresponding to each review), travel type (divided into single, family, couple, business, family, etc.), year-month fixed effects of the time of posting the review and hotel fixed effects; use the preprocessed data set obtained in step 2 to calculate the variables required for the research model (nonlinear regression model);

[0038] The variables of review usefulness, review text depth, rating, number of review images, management feedback, time interval, user experience, travel type, year-month fixed effect of review time, and hotel fixed effect are obtained from the TripAdvisor website. The three variables of review image aesthetics, review sentiment, and text readability need to be calculated with the help of tools; the calculation process is:

[0039] First, the aesthetics of the commented pictures is calculated using the multi-scale image quality transformer (MUSIQ) to calculate the aesthetics score of each picture. Then, the average aesthetics score of all pictures in each comment is used as a measure of the aesthetics of the commented pictures.

[0040] Secondly, the comment sentiment is measured by the following formula (1):

[0041]

[0042] Among them: i represents hotel, j represents comment, Emotion ij represents the sentiment score of the jth review of the i-th hotel, PW ij NW represents the number of positive sentiment words in the jth review of the i-th hotel. ij TW represents the number of negative sentiment words in the jth review of the i-th hotel. ijrepresents the total number of words in the jth review of the i-th hotel; the above sentiment word frequencies are obtained with the help of LIWC software, and the initial sentiment value scores of each review calculated according to formula (1) are between -1 and 1;

[0043] Finally, the text readability was calculated using the Gunning FOG index in Python software. The larger the Gunning FOG index, the worse the text readability;

[0044] Step 4: Nonlinear regression model analysis: Based on the research model (nonlinear regression model), the defined variables are set as explained variables, explanatory variables, moderating variables and control variables, specifically: the usefulness of the review is taken as the explained variable, the aesthetics of the review picture is taken as the explanatory variable, the depth of the review text is taken as the moderating variable, and the review sentiment, score, text readability, number of review pictures, management feedback, time interval, user experience, travel type, year-month fixed effects of time and hotel fixed effects are taken as control variables; a nonlinear regression model containing the above variables is constructed, and the construction process is as follows:

[0045] The explained variable is expressed as a function of the explanatory variable, the quadratic term of the explanatory variable, the constant term, the error term, the hotel fixed effect and the time fixed effect; the expression of the nonlinear regression model without a moderating variable is shown in the following formula (2):

[0046]

[0047] Where: i represents the hotel, j represents the review, Lg Helpfulness ij represents the logarithm of the number of helpfulness votes obtained by the jth review of the i-th hotel, Aesthetics ij Aesthetics represents the average aesthetic score of the pictures included in the jth review of the i-th hotel. 2 ij represents the quadratic term of the average aesthetic score of the pictures included in the jth review of the i-th hotel, Control ij represents the control variable for the jth review of the i-th hotel, λ ij The coefficient of the control variable for the jth review of the i-th hotel, Hotel i represents the fixed effect of the i-th hotel, y represents the year of the review, m represents the month of the review, and Time ym represents the year-month fixed effect of the time of posting comments, ε1 is a random disturbance term, β 10 represents the constant term, β 11 represents the linear influence of the aesthetics of the review image on the usefulness of the review, β 12 Indicates the nonlinear influence of the aesthetics of the review image on the usefulness of the review; focus on β11 and β 12 , if β 11 The regression coefficient is significant (the p-value is less than 0.1) and positive, and β 12 If the regression coefficient is negative (the p-value is less than 0.1), it means that the aesthetics of the review image has an inverted U-shaped effect on the usefulness of the review.

[0048] In order to further study the differences in the above relationship under different review text depth conditions, the review text depth is introduced into the model as a moderating variable. At this time, the explained variable is expressed as a function of the explanatory variable, the quadratic term of the explanatory variable, the moderating variable, the interaction term between the explanatory variable and the moderating variable, the interaction term between the quadratic term of the explanatory variable and the moderating variable, the constant term, the error term, the hotel fixed effect and the time fixed effect. The expression of the nonlinear regression model with the moderating variable is shown in the following formula (3):

[0049]

[0050] Where: i represents the hotel, j represents the review, Lg Helpfulness ij represents the logarithm of the number of helpfulness votes obtained by the jth review of the i-th hotel, Aesthetics ij Aesthetics represents the average aesthetic score of the pictures included in the jth review of the i-th hotel. 2 ij The quadratic term representing the average aesthetic score of the pictures included in the jth review of the i-th hotel, LgRevDepth ij represents the logarithm of the text depth of the jth review of the i-th hotel, Aesthetics ij ×LgRevDepth ij Aesthetics represents the intersection of the aesthetic score of the jth review image of the i-th hotel and the logarithm of the text depth. 2 ij ×LgRevDepth ij represents the intersection of the quadratic term of the aesthetic score of the jth review image of the i-th hotel and the logarithm of the text depth, Control ij represents the control variable for the jth review of the i-th hotel, λ ij The coefficient of the control variable for the jth review of the i-th hotel, Hotel i represents the fixed effect of the i-th hotel, y represents the year of the review, m represents the month of the review, and Time ym represents the year-month fixed effect of the time of posting comments, ε2 represents the random disturbance term, β 20 represents the constant term, β 21 represents the linear influence of the aesthetics of the review image on the usefulness of the review, β22 represents the nonlinear influence of the aesthetics of the review image on the usefulness of the review, β 23 represents the linear influence of review text depth on review usefulness, β 24 Indicates the influence of the intersection of review text depth and review image aesthetics on review usefulness, β 25 Indicates the influence of the intersection of the quadratic term of the review text depth and the review image aesthetics on the usefulness of the review; focus on β 25 , if β 25 If it is significant (the p-value of the regression coefficient is less than 0.1), it means that the depth of the comment text has a moderating effect;

[0051] Then, the variable data calculated in step 3 were imported into STATA software to fit the model, and the coefficients and residual terms of the nonlinear regression model were obtained (the coefficients can reveal the impact of the aesthetics of the review image on the usefulness of the review and the boundary conditions that affect the above relationship, while the residual term represents the unexplained part of the model);

[0052] Step 5: Robustness test; the nonlinear regression model in step 4 is tested for robustness by replacing the least squares regression method with the Tobit regression method and selecting a subsample set for testing; if the nonlinear regression model passes the robustness test, it means that the model is reliable, and then according to the coefficients and residual terms of the nonlinear regression model calculated in step 4, a prediction model of the usefulness of comments based on the aesthetics of the commented pictures is obtained; if it fails the robustness test, the nonlinear regression model needs to be rebuilt and analyzed, that is, return to execute step 4.

[0053] Embodiment 1:

[0054] This embodiment discloses a method for predicting the usefulness of reviews based on the aesthetics of review images. It uses data analysis to mine the review data of New York hotels in the international mainstream digital tourism platform TripAdvisor, defines and calculates key variables such as review usefulness, review image aesthetics, and review text depth, analyzes the impact of review image aesthetics and review usefulness by building a nonlinear regression model, and invents a method for predicting the usefulness of reviews based on review image aesthetics. The invention aims to predict the usefulness of reviews more accurately and comprehensively, and to help online travel platforms and hotel managers with online reputation management.

[0055] 1. Research Data and Methods

[0056] 1. Research data

[0057] The present invention selects user review data of New York hotels on the international mainstream online travel platform TripAdvisor.com as a research sample. The online travel platform integrates multiple service modules such as hotel accommodation, tourism, catering and food for global travelers, with more than 1 billion user reviews, covering nearly 8 million companies. The present invention designs a web crawler program based on Python language to obtain user review data from TripAdvisor.com, including usefulness voting, review rating, review text, review picture, review time, check-in time, whether management feedback is received, travel type, user experience and hotel ID. The present invention screens the review data containing at least one review photo, deletes missing values ​​and outliers, and finally obtains the preprocessed data set.

[0058] 2. Research Methods

[0059] The present invention proposes a method for predicting review usefulness based on review image aesthetics. Specifically, by performing data mining on review information (review usefulness, review images, and review text, etc.) on hotel websites, the nonlinear effect of review image aesthetics on review usefulness is analyzed.

[0060] like Figure 1 As shown, the present invention proposes a method for predicting the usefulness of comments based on the aesthetics of commented pictures, and the method comprises the following steps:

[0061] Step 1: Obtain the original data set; obtain the online review data of the target hotel (such as hotels in the New York area of ​​the United States) to obtain the original data set;

[0062] Step 2: Data preprocessing: Data preprocessing is performed on the original data set. First, the review data containing at least one review image is screened out, and then the samples with missing data and abnormal data are deleted to obtain the preprocessed data set (i.e., data screening and data cleaning).

[0063] Step 3: Variable definition and calculation: sort out and summarize the relevant literature in the field of online reviews, determine the research model (non-linear regression model) for the impact of review usefulness on review image aesthetics, define the variables required for the research model, and use the preprocessed data set obtained in step 2 to calculate the variables required for the research model (non-linear regression model);

[0064] Step 4: Nonlinear regression model analysis: Based on the research model (nonlinear regression model), the defined variables are set as explained variables, explanatory variables, moderating variables and control variables, and a nonlinear regression model containing the above variables is constructed. Then, the variable data obtained by calculation in step 3 is imported into STATA software to fit the model, and the coefficients and residual terms of the nonlinear regression model are obtained (the coefficients can reveal the impact of the aesthetics of the review picture on the usefulness of the review and the boundary conditions that affect the above relationship, while the residual term represents the unexplained part of the model);

[0065] Step 5: Robustness test; the nonlinear regression model in step 4 is tested for robustness by replacing the least squares regression method with the Tobit regression method and selecting a subsample set for testing; if the nonlinear regression model passes the robustness test, it means that the model is reliable, and then according to the coefficients and residual terms of the nonlinear regression model calculated in step 4, a prediction model of the usefulness of comments based on the aesthetics of the commented pictures is obtained; if it fails the robustness test, the nonlinear regression model needs to be rebuilt and analyzed, that is, return to execute step 4.

[0066] 2. Experiment and Analysis

[0067] 1. Data source and preprocessing

[0068] The present invention uses a Python crawler program to obtain user review data of all hotels in New York, USA from TripAdvisor.com, including usefulness votes, review scores, review text, review pictures, review time, check-in time, whether management feedback has been received, travel type, user experience, and hotel ID. The present invention screens review data containing at least one review photo, deletes missing values ​​and outliers, and finally obtains a data set containing 55,094 user reviews. This embodiment analyzes based on such review data on this platform.

[0069] 2. Variable measurement

[0070] There are many types of data obtained from TripAdvisor.com, and many of them cannot be directly used for analysis. Therefore, it is necessary to measure the relevant variables. The specific variable definition and measurement process are as follows:

[0071] (1) Explained variable: The number of helpfulness votes each comment received (Helpfulness), which is a common indicator for measuring the helpfulness of comments.

[0072] (2) Explanatory variables: the average aesthetic score of all review images contained in each review (Aesthetics). The specific operation process of this variable is as follows: First, download the review images of each review. Second, use the multi-scale image quality transformer (MUSIQ) to calculate the aesthetic score of each image. Finally, use the average aesthetic score of all images in each review as a measure to aggregate the data to the review level.

[0073] (3) Moderating variable: Review depth (RevDepth). The present invention uses the number of words in a review to measure the depth of the review.

[0074] (4) Control variables: We controlled for variables related to reviews, including the difference in the proportion of positive and negative emotion-related words expressed in each review (Emotion), the total score of each review (Rating), the readability of each review text (Readability), the total number of pictures contained in each review (RevPicNum), whether the review received management feedback (HaveResponse), the interval between each review's check-in and the posting of the review (Interval), the experience value of the user account corresponding to each review (UserExperience), the type of trip (TripType), the year-month fixed effect of the time when the review was posted, and the hotel fixed effect. Among them, the variables such as rating, number of review pictures, management feedback, time interval, user experience, trip type, year-month fixed effect of the time when the review was posted, and hotel fixed effect can be obtained from the website, while the review emotion and text readability need to be calculated with the help of tools.

[0075] First, the comment sentiment is measured by the following formula (1):

[0076]

[0077] Among them: i represents hotel, j represents comment, Emotion ij represents the sentiment score of the jth review of the i-th hotel, PW ij represents the number of positive sentiment words in the jth review of the i-th hotel, NW ij represents the number of negative sentiment words in the jth review of the i-th hotel, TW ij represents the total number of words in the jth review of the i-th hotel; the above sentiment word frequencies are obtained with the help of LIWC software, and the initial sentiment value scores of each review calculated according to formula (1) are between -1 and 1;

[0078] Secondly, the text readability was calculated using the Gunning FOG index in Python software. The larger the Gunning FOG index, the worse the text readability.

[0079] Due to the high standard deviation and skewness, we performed logarithmic transformation on the five variables, Helpfulness, RevDepth, RevPicNum, Interval, and UserExperience, and used LgHelpfulness, LgRevDepth, LgRevPicNum, LgInterval, and LgUserExperience to represent the logarithm of the number of helpfulness votes obtained, the logarithm of the review text depth, the logarithm of the number of pictures contained in the reviews, the logarithm of the time interval between reviews, and the logarithm of the user’s account experience, respectively.

[0080] 3. Nonlinear regression model analysis

[0081] Based on the research model, the usefulness of reviews is taken as the explained variable, the aesthetics of review pictures is taken as the explanatory variable, the depth of review text is taken as the moderating variable, and the review sentiment, rating, text readability, number of review pictures, management feedback, time interval, user experience, travel type, year-month fixed effects of time and hotel fixed effects are taken as control variables to construct a nonlinear regression model containing these variables. The results show that the aesthetics of review pictures has an inverted U-shaped effect on the usefulness of reviews, and this relationship is more significant when the depth of review text is low.

[0082] 3.1 Model Building

[0083] First, a nonlinear regression model of the effect of the aesthetics of review images on the usefulness of reviews is constructed. Specifically, the explained variable is expressed as a function of the explanatory variable, the quadratic term of the explanatory variable, the constant term, the error term, the hotel fixed effect, and the time fixed effect; the expression of the nonlinear regression model without a moderating variable is shown in the following formula (2):

[0084]

[0085] Where: i represents the hotel, j represents the review, Lg Helpfulness ij represents the logarithm of the number of helpfulness votes obtained by the jth review of the i-th hotel, Aesthetics ij Aesthetics represents the average aesthetic score of the pictures included in the jth review of the i-th hotel. 2 ij represents the quadratic term of the average aesthetic score of the pictures included in the jth review of the i-th hotel, Control ij represents the control variable for the jth review of the i-th hotel, λ ijThe coefficient of the control variable for the jth review of the i-th hotel, Hotel i represents the fixed effect of the i-th hotel, y represents the year of the review, m represents the month of the review, and Time ym represents the year-month fixed effect of the time of posting comments, ε1 is a random disturbance term, β 10 represents the constant term, β 11 represents the linear influence of the aesthetics of the review image on the usefulness of the review, β 12 Indicates the nonlinear influence of the aesthetics of the review image on the usefulness of the review; focus on β 11 and β 12 , if β 11 The regression coefficient is significant (the p-value is less than 0.1) and positive, and β 12 If it is significant (the p-value of the regression coefficient is less than 0.1) and negative, it means that the aesthetics of the review image has an inverted U-shaped effect on the usefulness of the review;

[0086] Secondly, in order to further study the differences in the above relationship under different review text depth conditions, the present invention introduces the review text depth as a moderating variable into the model. At this time, the explained variable is expressed as a function of the explanatory variable, the quadratic term of the explanatory variable, the moderating variable, the interaction term between the explanatory variable and the moderating variable, the interaction term between the quadratic term of the explanatory variable and the moderating variable, the constant term, the error term, the hotel fixed effect and the time fixed effect; the expression of the nonlinear regression model with the moderating variable is shown in the following formula (3):

[0087]

[0088] Where: i represents the hotel, j represents the review, Lg Helpfulness ij represents the logarithm of the number of helpfulness votes obtained by the jth review of the i-th hotel, Aesthetics ij Aesthetics represents the average aesthetic score of the pictures included in the jth review of the i-th hotel. 2 ij The quadratic term representing the average aesthetic score of the pictures included in the jth review of the i-th hotel, LgRevDepth ij represents the logarithm of the text depth of the jth review of the i-th hotel, Aesthetics ij ×LgRevDepth ij Aesthetics represents the intersection of the aesthetic score of the jth review image of the i-th hotel and the logarithm of the text depth. 2 ij ×LgRevDepth ij represents the intersection of the quadratic term of the aesthetic score of the jth review image of the i-th hotel and the logarithm of the text depth, Controlij represents the control variable for the jth review of the i-th hotel, λ ij The coefficient of the control variable for the jth review of the i-th hotel, Hotel i represents the fixed effect of the i-th hotel, y represents the year of the review, m represents the month of the review, and Time ym represents the year-month fixed effect of the time of posting comments, ε2 represents the random disturbance term, β 20 represents the constant term, β 21 represents the linear influence of the aesthetics of the review image on the usefulness of the review, β 22 represents the nonlinear influence of the aesthetics of the review image on the usefulness of the review, β 23 represents the linear influence of review text depth on review usefulness, β 24 Indicates the influence of the intersection of review text depth and review image aesthetics on review usefulness, β 25 Indicates the influence of the intersection of the quadratic term of the review text depth and the review image aesthetics on the usefulness of the review; focus on β 25 , if β 25 If the regression coefficient is significant (the p-value of the regression coefficient is less than 0.1), it means that the depth of the comment text has a moderating effect.

[0089] 3.2 Calculation of model coefficients and residual terms

[0090] In this embodiment, Stata software is used to run the above model, and the output results are shown in Table 1. The results of model (1) show that the coefficient between the review image aesthetics and the review usefulness is significantly positive, while the coefficient between the quadratic term of the review image aesthetics and the review usefulness is significantly negative. This shows that there is an inverted U-shaped relationship between the review image aesthetics and the review usefulness. The results of model (2) show that the intersection term of the review image aesthetics and the review text depth has a significant negative impact on the review usefulness, while the intersection term of the quadratic term of the review image aesthetics and the review text depth has a significant positive impact on the review usefulness. This shows that the review text depth plays a weakening moderating role in the relationship between the review image aesthetics and the review usefulness. In other words, the inverted U-shaped relationship between the review image aesthetics and the review usefulness is more significant when the review text depth is low. At the same time, the R of model (1) and model (2) is 2 (coefficient of determination) are 0.2571 and 0.2597 respectively, indicating that all variables can explain about 26% of the variation of the dependent variable, and the fitting effect is good.

[0091] Table 1

[0092]

[0093]

[0094] Note: The numbers in brackets in the table are standard errors of the variables; ***p<0.01, **p<0.05, *p<0.1.

[0095] 4. Robustness test

[0096] 4.1 Alternative regression method

[0097] In the present invention, 88.39% of the dependent variable values ​​(i.e., the usefulness vote of each review) are zero. This embodiment uses the Tobit regression method to perform a robustness test, and the results are shown in Table 2. This result is consistent with the previous results, which improves the reliability of this prediction method.

[0098] Table 2

[0099]

[0100]

[0101] Note: The numbers in brackets in the table are standard errors of the variables; ***p<0.01, **p<0.05, *p<0.1.

[0102] 4.2 Sub-sampling with replacement

[0103] It is voluntary for users to write reviews, so there is no time limit. However, if the time interval between checking in to a hotel and posting a review is long, there may be a significant recall bias, which will affect the accuracy of the review. Therefore, the embodiment excludes reviews with a long time interval. Table 3 shows the OLS regression and Tobit regression results for reviews with a time interval of two months and three months. The results remain consistent, further strengthening the robustness of this prediction method.

[0104] Table 3

[0105]

[0106]

[0107] Note: The numbers in brackets in the table are standard errors of the variables; ***p<0.01, **p<0.05, *p<0.1.

[0108] The nonlinear regression models with (or without) moderating variables have demonstrated their reliability in different data distributions and noise environments through robustness tests. According to the actual case, you can choose a nonlinear regression model with or without moderating variables: if the case has no moderating variables, you should choose a nonlinear regression model without moderating variables, and bring the coefficients of the explanatory variable (the aesthetics of the review picture), the quadratic term of the explanatory variable (the quadratic term of the aesthetics of the review picture), and the control variables (review sentiment, rating, text readability, number of review pictures, management feedback, time interval, user experience, travel type, time fixed effect, hotel fixed effect) into the nonlinear regression model without moderating variables to get a prediction model of review usefulness based on review picture aesthetics; if the case has a moderating variable, you should choose a nonlinear regression model with ... The coefficients of the explanatory variable (aesthetics of review pictures), the quadratic term of the explanatory variable (quadratic term of aesthetics of review pictures), the moderating variable (depth of review text), the intersection term of the explanatory variable and the moderating variable (aesthetics of review pictures × depth of review text), the intersection term of the quadratic term of the explanatory variable and the moderating variable (quadratic term of aesthetics of review pictures × depth of review text), and the control variables (review sentiment, rating, text readability, number of review pictures, management feedback, time interval, user experience, travel type, time fixed effects, and hotel fixed effects) are substituted into the nonlinear regression model with the moderating variable to obtain a prediction model for review usefulness based on review picture aesthetics.

[0109] The above are only preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any equivalent replacement or modification made by any technician familiar with the technical field within the technical scope disclosed by the present invention and based on the technical solution and concept of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting the usefulness of reviews based on the aesthetics of reviewed images, characterized by: The method comprises the following steps: Step 1: Obtain the original data set; obtain the online review data of the target hotel to obtain the original data set; Step 2: Data preprocessing: Perform data preprocessing on the original data set. First, filter out the review data that contains at least one review image, then delete the samples with missing data and abnormal data to obtain the preprocessed data set. Step 3: Variable definition and calculation: sort out and summarize the relevant literature in the field of online reviews, determine the research model of the impact of review usefulness on review image aesthetics, define the variables required for the research model, and use the preprocessed data set obtained in step 2 to calculate the variables required for the research model; Step 4: Nonlinear regression model analysis: Based on the research model, the defined variables are set as explained variables, explanatory variables, moderating variables and control variables, and a nonlinear regression model containing the above variables is constructed. Then, the variable data obtained by calculation in step 3 is imported into STATA software to fit the model, and the coefficients and residual terms of the nonlinear regression model are obtained. Step 5: Robustness test; the nonlinear regression model in step 4 is tested for robustness by replacing the least squares regression method with the Tobit regression method and selecting a subsample set for testing; if the nonlinear regression model passes the robustness test, it means that the model is reliable, and then according to the coefficients and residual terms of the nonlinear regression model calculated in step 4, a prediction model of the usefulness of comments based on the aesthetics of the commented pictures is obtained; if it fails the robustness test, the nonlinear regression model needs to be rebuilt and analyzed, that is, return to execute step 4.

2. The method for predicting review usefulness based on review image aesthetics according to claim 1, characterized in that: In step 1, the online review data of the target hotel is obtained from the TripAdvisor website through a crawler program.

3. The method for predicting review usefulness based on review image aesthetics according to claim 2, characterized in that: The crawler program is used to obtain online review data of the target hotel from the TripAdvisor website, including usefulness votes, review scores, review text, review pictures, review time, check-in time, whether management feedback has been received, travel type, user experience and hotel ID.

4. The method for predicting review usefulness based on review image aesthetics according to claim 1, characterized in that: In step 3, the variables required for the research model were defined, including review helpfulness, review image aesthetics, review text depth, review sentiment, rating, text readability, number of review images, management feedback, time interval, user experience, travel type, year-month fixed effects of the time when the review was posted, and hotel fixed effects.

5. The method for predicting review usefulness based on review image aesthetics according to claim 4, characterized in that: The variables of review usefulness, review text depth, rating, number of review images, management feedback, time interval, user experience, travel type, year-month fixed effect of review time, and hotel fixed effect are obtained from the TripAdvisor website. The three variables of review image aesthetics, review sentiment, and text readability need to be calculated with the help of tools; the calculation process is: First, the aesthetics of the commented pictures variable is calculated using a multi-scale image quality converter to calculate the aesthetics score of each picture. Then, the average aesthetics score of all pictures in each comment is used as a measure of the aesthetics of the commented pictures. Secondly, the comment sentiment is measured by the following formula (1): Among them: i represents hotel, j represents comment, Emotion ij represents the sentiment score of the jth review of the i-th hotel, PW ij NW represents the number of positive sentiment words in the jth review of the i-th hotel. ij TW represents the number of negative sentiment words in the jth review of the i-th hotel. ij represents the total number of words in the jth review of the i-th hotel; the above sentiment word frequencies are obtained with the help of LIWC software, and the initial sentiment value scores of each review calculated according to formula (1) are between -1 and 1; Finally, the text readability was calculated using the Gunning FOG index in Python software. The larger the Gunning FOG index, the worse the text readability.

6. The method for predicting review usefulness based on review image aesthetics according to claim 1, characterized in that: In step 4, the defined variables are set as explained variables, explanatory variables, adjustment variables and control variables respectively, specifically: The helpfulness of reviews was taken as the explained variable, the aesthetics of review pictures as the explanatory variable, the depth of review text as the moderating variable, and review sentiment, rating, text readability, number of review pictures, management feedback, time interval, user experience, travel type, year-month fixed effects of time, and hotel fixed effects as control variables.

7. The method for predicting review usefulness based on review image aesthetics according to claim 6, characterized in that: In step 4, the nonlinear regression model construction process is: the explained variable is expressed as a function of the explanatory variable, the quadratic term of the explanatory variable, the constant term, the error term, the hotel fixed effect and the time fixed effect; the expression of the nonlinear regression model without a moderating variable is shown in the following formula (2): Where: i represents the hotel, j represents the review, Lg Helpfulness ij represents the logarithm of the number of helpfulness votes obtained by the jth review of the i-th hotel, Aesthetics ij Aesthetics represents the average aesthetic score of the pictures included in the jth review of the i-th hotel. 2 ij represents the quadratic term of the average aesthetic score of the pictures included in the jth review of the i-th hotel, Control ij represents the control variable for the jth review of the i-th hotel, λ ij The coefficient of the control variable for the jth review of the i-th hotel, Hotel i represents the fixed effect of the i-th hotel, y represents the year of the review, m represents the month of the review, and Time ym represents the year-month fixed effect of the time of posting comments, ε1 is a random disturbance term, β 10 represents the constant term, β 11 represents the linear influence of the aesthetics of the review image on the usefulness of the review, β 12 Indicates the nonlinear influence of the aesthetics of the review image on the usefulness of the review; focus on β 11 and β 12 , if β 11 is significantly positive, and β 12 If it is significantly negative, it means that the aesthetics of the review image has an inverted U-shaped effect on the usefulness of the review.

8. The method for predicting review usefulness based on review image aesthetics according to claim 6, characterized in that: The review text depth is introduced into the model as a moderating variable. At this time, the explained variable is expressed as a function of the explanatory variable, the quadratic term of the explanatory variable, the moderating variable, the interaction term between the explanatory variable and the moderating variable, the interaction term between the quadratic term of the explanatory variable and the moderating variable, the constant term, the error term, the hotel fixed effect and the time fixed effect. The expression of the nonlinear regression model with the moderating variable is shown in the following formula (3): Where: i represents the hotel, j represents the review, Lg Helpfulness ij represents the logarithm of the number of helpfulness votes obtained by the jth review of the i-th hotel, Aesthetics ij Aesthetics represents the average aesthetic score of the pictures included in the jth review of the i-th hotel. 2 ij The quadratic term representing the average aesthetic score of the pictures included in the jth review of the i-th hotel, LgRevDepth ij represents the logarithm of the text depth of the jth review of the i-th hotel, Aesthetics ij ×LgRevDepth ij Aesthetics represents the intersection of the aesthetic score of the jth review image of the i-th hotel and the logarithm of the text depth. 2 ij ×LgRevDepth ij represents the intersection of the quadratic term of the aesthetic score of the jth review image of the i-th hotel and the logarithm of the text depth, Control ij represents the control variable for the jth review of the i-th hotel, λ ij The coefficient of the control variable for the jth review of the i-th hotel, Hotel i represents the fixed effect of the i-th hotel, y represents the year of the review, m represents the month of the review, and Time ym represents the year-month fixed effect of the time of posting comments, ε2 represents the random disturbance term, β 20 represents the constant term, β 21 represents the linear influence of the aesthetics of the review image on the usefulness of the review, β 22 represents the nonlinear influence of the aesthetics of the review image on the usefulness of the review, β 23 represents the linear influence of review text depth on review usefulness, β 24 Indicates the influence of the intersection of review text depth and review image aesthetics on review usefulness, β 25 Indicates the influence of the intersection of the quadratic term of the review text depth and the review image aesthetics on the usefulness of the review; focus on β 25 , if β 25 If it is significant, it means that the depth of comment text has a moderating effect.

Citation Information

Patent Citations

  • Non-local language customer satisfaction prediction method based on hotel service language

    CN117094855A

  • Method for predicting emotions of consumers in emergent public health event

    CN117114746A

  • A computer-implemented method and system for analyzing and evaluating user reviews

    US20180260860A1