A correction method to remove the interference of online review presentation order on review helpfulness

Abstract

The present invention discloses a correction method for removing the interference of the presentation order of online comments on the helpfulness of comments, by constructing a panel data set including the presentation order and the helpfulness of comments, and assigning values; collecting multi-period cross-sectional data and making comment associations for the same comments; calculating the newly added votes of the comments to form a panel data set; calculating the comment retention rate and the order stability rate; judging whether the comment retention rate and the order stability rate meet the requirements; calculating the coefficient of the comment presentation order; and correcting the error of the comment helpfulness according to the influence of the order bias. The present invention is suitable for improving the analysis of the helpfulness of comments on online platforms. By analyzing the information of the presentation order, the system bias existing in the helpfulness of comments from viewers is quantitatively characterized, and high-quality comments are effectively identified through error correction, thereby reducing the information overload of consumers.

Description

Technical Field

[0001] The invention belongs to the technical field of data processing, and in particular relates to a correction method for removing interference of online comment presentation order on comment helpfulness. Background Art

[0002] With the development of the e-commerce market, online platforms are paying more and more attention to the role of online reviews in helping consumers make purchases. In order to reduce the information overload brought to consumers by massive review information, the platform analyzes the quality of reviews by collecting votes from viewers and gives priority to presenting high-quality reviews to consumers. However, this information from viewers' votes cannot fully reflect the quality of reviews. Since the visibility of comments ranked lower decreases and the marginal utility of information decreases, the voting information received by the comments will be negatively affected by the presentation order, resulting in errors in the quality analysis of reviews based on voting information.

[0003] Therefore, a technology is needed that can remove the bias caused by the presentation order based on the voting information, so as to help the platform more accurately estimate the quality of comments. Summary of the invention

[0004] In order to solve the above-mentioned defects existing in the prior art, the purpose of the present invention is to provide a correction method for removing the interference of the presentation order of online comments on the helpfulness of comments. By constructing a panel data set including the presentation order of comments and the helpfulness of comments, the impact of the comment order bias is calculated, and the error correction of the comment helpfulness is performed, thereby helping online platforms to more accurately estimate the quality of comments.

[0005] The present invention is achieved through the following technical solutions.

[0006] In one aspect, the present invention provides a method for correcting the interference of the presentation order of online comments on the helpfulness of comments, comprising:

[0007] Construct a panel data set that includes the order of review presentation and review helpfulness;

[0008] The comment collection window and comment collection interval are respectively set, and values ​​are assigned to the comment collection window and comment collection interval;

[0009] Conduct multi-period cross-sectional data collection;

[0010] Comments of the same comments in two adjacent cross-sectional data periods were linked;

[0011] Based on the obtained correlation between comments in two consecutive periods, the newly added votes of the comments are calculated to form a panel data set;

[0012] Using the panel data set, the review retention rate and sequential stability rate are calculated;

[0013] Determine whether the comment retention rate and order stability rate meet the requirements. If not, reassign the collection interval and collection window; otherwise, calculate the coefficient of the comment presentation order;

[0014] For the obtained panel data set, the coefficients of the order in which reviews are presented are calculated;

[0015] The helpfulness of comments is adjusted using the comment presentation order coefficient to obtain the corrected helpfulness of comments.

[0016] Preferably, the comment collection window and the comment collection interval are assigned values, and the first initial value is preset according to the requirements of the engineering analysis; if the comment retention rate obtained last time does not meet the requirements, the collection window is expanded; if the sequence stability rate obtained last time does not meet the requirements, the collection interval is reduced; the expansion amount of the collection window and the reduction amount of the collection interval are preset according to the requirements of the engineering analysis.

[0017] Preferably, cross-sectional data collection includes three dimensions: product, comment and issue; the collected data includes four types: total number of votes, presentation order, comment text and control variables; the total number of issues for data collection is preset according to engineering analysis requirements; the variables included in the control variables are preset according to engineering analysis requirements.

[0018] Preferably, associating two adjacent periods of comments refers to identifying identical comments in two adjacent periods of comment sets based on the comment texts.

[0019] Preferably, the requirement for the comment retention rate and the sequence stability rate is to be greater than a given threshold, and the given threshold of the comment retention rate and the sequence stability rate is preset according to the requirements of the engineering analysis.

[0020] Preferably, for the obtained panel data set Ω, the coefficient of the review presentation order is calculated using least squares regression.

[0021] The present invention adopts the above technical solution, which has the following beneficial effects:

[0022] 1. Reduce information overload for consumers. The present invention uses the obtained panel data set to calculate the coefficient of the order of comment presentation, thereby achieving a quantitative characterization of the systematic bias between the helpfulness of comments voted by viewers and the true quality of the comments.

[0023] 2. The error correction method of adjusting the helpfulness of reviews by using the review presentation order coefficient achieves effective identification of high-quality reviews, thereby reducing consumers' information overload.

[0024] 3. It has strong analytical level transferability. This method can be applied to the platform-level review presentation order to adjust the bias of its helpfulness, and can also be applied at the product category level and product level, with flexible transferability. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The drawings described herein are used to provide a further understanding of the present invention, constitute a part of the present application, and do not constitute an improper limitation of the present invention. In the drawings:

[0026] Figure 1 This is a flow chart of a correction algorithm for removing helpful interference of presentation order on online reviews according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments. The exemplary embodiments and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0028] Figure 1 This is a flow chart of a correction method for removing the interference of the online comment presentation order on the helpfulness of comments according to an embodiment of the present invention. By constructing a panel data set containing the comment presentation order and comment helpfulness, the influence of comment order bias is calculated to achieve error correction of comment helpfulness. Comment helpfulness is the number of likes a comment receives from viewers; comment presentation order is the order variable of a comment presented to viewers on a page; and order bias is the influence of comment presentation order on the number of likes a comment receives.

[0029] The deviation correction method of the present invention comprises the following steps:

[0030] Step 1: construct a panel dataset containing the order of review presentation and review helpfulness.

[0031] First, the comment collection window and comment collection interval are set respectively, and values ​​are assigned to the comment collection window and comment collection interval.

[0032] The first time the initial value is assigned, the value is preset according to the requirements of the engineering analysis; during the iterative solution, the comment retention rate and the sequence stability rate calculated in the previous step 5 are corrected.

[0033] The collection window and collection interval are adjusted during iterations as follows: if the comment retention rate obtained last time according to step 6 does not meet the requirements, the collection window is expanded; if the sequence stability rate obtained last time according to step 6 does not meet the requirements, the collection interval is reduced; the amount of expansion of the collection window and the amount of reduction of the collection interval are preset according to the requirements of engineering analysis.

[0034] Step 2: Collect multi-period cross-sectional data according to the collection window and collection interval set in step 1.

[0035] Cross-sectional data collection includes three dimensions: {i, j, t}, where i = 1, 2, ..., I represents comments, I is the total number of comments on a product in the collection window; j = 1, 2, ..., J represents products, J is the total number of products collected, which is preset according to the requirements of engineering analysis; t = 1, 2, ..., T represents the number of periods, T is the total number of periods of cross-sectional data collected, which is preset according to the requirements of engineering analysis.

[0036] The collected data includes the following four types: ijt ,order ijt ,text ij ,controls ijt}, where votes ijt is the total number of votes received by review i of product j as of period t, which is an ordinal variable; order ijt is the order in which the review i of product j is presented in period t, which is an ordinal variable; text ij is the text content of the review i of product j, which is a string variable; controls ijt It is the control variable of review i of product j in period t, which can contain multiple numerical variables and is preset according to the requirements of engineering analysis.

[0037] Step 3: Based on the obtained multi-period cross-sectional data, the comments of two adjacent cross-sectional data are associated. The comment sets of two adjacent periods are associated based on the comment text text ij Identify identical comments.

[0038] Step 4: Based on the obtained correlation between comments in two adjacent periods, the newly added votes of the comments are calculated to construct a panel data set.

[0039] Using the correlation between comments in two consecutive periods, calculate the new votes for the comments: y ijt =votes ij(t+1) -votes ijt , forming a panel data set Ω;

[0040] Each observation in the panel data set Ω contains the following five main types:

[0041] {y ijt ,order ijt ,votes ijt ,text ijt ,controls ijt}, where t=1,2,…T-1.

[0042] Step 5, calculate the comment retention rate and sequence stability rate based on the panel data set.

[0043] The formula for calculating comment retention rate is: Where N is the total number of observations in the panel dataset Ω, and M is the number of reviews in the panel dataset Ω.

[0044] The sequential stability rate calculation formula is: Where N stable is the order of two consecutive periods in the panel data set Ω ijt The number of observations that did not change, N 2,3,...T-1 is the number of observations t∈{2,3,…T-1} in the panel dataset Ω.

[0045] Step 6: Determine the comment retention rate and sequence stability rate.

[0046] Determine whether the comment retention rate and sequence stability rate meet the requirements. If not, return to step 1, otherwise go to step 7.

[0047] The requirement for the comment retention rate and the sequence stability rate is that they are greater than a given threshold, and the given thresholds for the comment retention rate and the sequence stability rate are preset according to the requirements of engineering analysis.

[0048] Step 7: Calculate the impact of order bias

[0049] For the panel data set Ω, the coefficient of presentation order is calculated using least squares regression; the regression analysis uses the following formula:

[0050] y ijt =α0+α1×ln(order ijt )+β×controls ijt +μ j +ν t +ε ijt ,

[0051] Among them, α0 is the constant term of regression, α1 is the coefficient of the order of comment presentation after taking the natural logarithm, β is the coefficient of the control variable, μ j is the fixed effect of the product, ν t is the fixed effect of the period, ε ijt is the regression-specific error term.

[0052] For the panel data set Ω, the least squares method is used to estimate the parameters according to the regression equation to obtain α1.

[0053] Step 8: Comments to help correct mistakes

[0054] Use the obtained presentation order coefficient α1 to adjust the helpfulness of the comments to obtain the corrected helpfulness of the comments. The correction uses the following formula: adjusted helpfulness ijt =y ijt -α1×ln(order ijt).

[0055] Step 9: Output the helpfulness of the comments after correction.

[0056] The implementation process of the present invention is further described below through a specific embodiment.

[0057] This example selects 660 reviews of high-selling products in six product categories of digital, computer, mobile phone, men's clothing, women's clothing, and beauty products on a large e-commerce platform as the research object. The following is a demonstration of the application of the method of the present invention on an online platform in the order of the above 9 steps.

[0058] According to the requirements of step 1, set the comment collection window to the first 50 comments in the comment area, and set the comment collection interval to 1 day.

[0059] Cross-sectional data were collected according to the requirements of step 2. The total number of periods for cross-sectional data collection was set to 64.

[0060] According to steps 3 and 4, the comments of two adjacent cross-sectional data are associated, and the new votes corresponding to the comments are calculated to obtain the panel data set Ω. The panel data set Ω contains data of 40,622 different comments in 63 periods, totaling 1,935,179 observations.

[0061] Calculate and judge the comment retention rate and order stability rate according to the requirements of steps 5 and 6. Set the threshold of comment retention rate and order stability rate to 60%.

[0062]

[0063] Here, the comment retention rate and sequential stability rate under the panel data set Ω are both greater than the given threshold, so we proceed to step 7.

[0064] Regress the panel data set Ω according to the requirements of step 7. The regression uses the following formula:

[0065] helpfulness ijt =α0+α1ln(order ijt )+β1votes ijt +β2controls ijt +μ j +ν t +ε ijt ,

[0066] The variables involved in the regression, their meanings and related statistical values ​​are shown in Table 1.

[0067] Table 1 Descriptive statistics of regression variables

[0068]

[0069]

[0070] The regression results obtained using the least squares method are shown in Table 2.

[0071] Table 2 Regression results

[0072]

[0073] Note: The robust t-statistics for coefficient estimates are in parentheses

[0074] ***p<0.01,**p<0.05,*p<0.1

[0075] We obtain α1 = -0.0288.

[0076] According to the requirements of steps 8 to 9, use the following formula to calculate the helpfulness of the corrected comments and output them:

[0077] adjusted helpfulness ijt =y ijt +0.0288×ln(order ijt ).

[0078] In addition, this method can be applied to the helpfulness correction of reviews at the product category level. The implementation method is to use the regression formula to perform sub-sample regression on the data of different product categories to obtain α1 under different product categories. The regression results under different product categories are shown in Table 3. In order to present the results concisely, the coefficient estimation results of control variables are omitted here.

[0079] Table 3 Regression results for different product categories

[0080]

[0081] Note: The robust t-statistics for coefficient estimates are in parentheses

[0082] ***p<0.01,**p<0.05,*p<0.1

[0083] From the above examples, it can be seen that the method of the present invention can effectively capture the error α1 caused by the presentation order from the product review information. In addition, the method of the present invention can be flexibly transferred to the product category level, so as to perform more sophisticated review-helpful correction.

[0084] The present invention is not limited to the above-mentioned embodiments. On the basis of the technical solution disclosed in the present invention, technicians in this field can make some substitutions and deformations to some technical features therein according to the disclosed technical content without creative labor, and these substitutions and deformations are all within the protection scope of the present invention.

Claims

1. A correction method for removing the interference of online review presentation order on review helpfulness, characterized in that: include: Construct a panel data set that includes review presentation order and review helpfulness; The comment collection window and comment collection interval are respectively set, and values ​​are assigned to the comment collection window and comment collection interval; Conduct multi-period cross-sectional data collection; Comments of the same comments in two adjacent cross-sectional data periods were linked; Based on the obtained correlation between comments in two consecutive periods, the newly added votes of the comments are calculated to form a panel data set; Calculate the new votes y for the comment ijt , the panel data set Ω is calculated according to the following formula: and ijt =votes ij(t+1) -votes ijt Each observation in the panel data set Ω includes: {votes ijt ,order ijt ,text ij ,controls ijt }, where t = 1, 2, ... T-1; Among them, votes ijt is the total number of votes received by review i of product j as of period t, which is an ordinal variable; order ijt is the order in which the review i of product j is presented in period t, which is an ordinal variable; text ij is the text content of the review i of product j, which is a string variable; controls ijt is the control variable for review i of product j in period t; Using the panel data set, we calculate the review retention rate and sequential stability rate; Calculate the comment retention rate and sequence stability rate using the following formula: Comment retention rate: Where N is the total number of observations in the panel data set Ω, and M is the number of reviews in the panel data set Ω; Sequential stability rate: Among them, N stable is the order of two consecutive periods in the panel data set Ω ijt The number of observations that did not change, N 2,3,...T-1 is the number of observations t∈{2,3,…T-1} in the panel data set Ω; Determine whether the comment retention rate and order stability rate meet the requirements. If not, reassign the collection interval and collection window; otherwise, calculate the coefficient of the comment presentation order; For the obtained panel data set, the coefficients of the order in which reviews are presented are calculated; For the obtained panel data set Ω, the coefficient of the order of review presentation is calculated using least squares regression: y ijt =α0+α1×ln(order ijt )+β×controls ijt +m j +v t +e ijt Among them, α0 is the constant term of regression, α1 is the coefficient of the order of comment presentation after taking the natural logarithm, β is the coefficient of the control variable, μ j is the fixed effect of the product, ν t is the fixed effect of the period, ε ijt is the regression-specific error term; The helpfulness of comments is adjusted using the comment presentation order coefficient to obtain the corrected helpfulness of comments. The helpfulness correction of comments is calculated using the following formula: adjustedhelpfulness ijt =y ijt -α1×ln(order ijt )。 2. A method for correcting the interference of online review presentation order on review helpfulness according to claim 1, characterized in that: The comment collection window and comment collection interval are assigned values. The first initial value is preset according to the requirements of engineering analysis. If the comment retention rate obtained last time does not meet the requirements, the collection window is expanded. If the sequence stability rate obtained last time does not meet the requirements, the collection interval is reduced. The expansion amount of the collection window and the reduction amount of the collection interval are preset according to the requirements of engineering analysis.

3. A method for correcting the interference of online review presentation order on review helpfulness according to claim 1, characterized in that: Cross-sectional data collection includes three dimensions: product, comment and issue; the collected data includes four types: total number of votes, presentation order, comment text and control variables; the total number of issues for data collection is preset according to the requirements of engineering analysis; the variables included in the control variables are preset according to the requirements of engineering analysis.

4. A method for correcting the interference of online review presentation order on review helpfulness according to claim 1, characterized in that: The association of two adjacent review periods refers to the identification of identical reviews in two adjacent review sets based on the review texts.

5. A method for removing the interference of online review presentation order on review helpfulness according to claim 1, characterized in that: The requirement for the comment retention rate and the sequence stability rate is that they are greater than a given threshold, and the given thresholds for the comment retention rate and the sequence stability rate are preset according to the requirements of engineering analysis.

Images