Description of information processing method, device, apparatus and computer readable storage medium

Abstract

The present invention provides a description information processing method, device, equipment and computer-readable storage medium. The method of the present invention determines the effective description of the target item and its weight information by preprocessing the description information of the target item; extracts the description feature data of the effective description of the target item; inputs the weight information and description feature data of the effective description of the target item into the description quality prediction model, and calculates the description quality information of the effective description of the target item, which can fully explore the effective features of the description information, integrate the description information features of multiple dimensions, accurately predict the description quality information of the effective description of the target item, sort the effective description of the target item according to the description quality information of the effective description of the target item, and realize accurate sorting of the effective description of the item. On this basis, the effective description of the target item can be displayed to the user in the order after re-sorting, which can improve the order conversion rate brought by the description information.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a description information processing method, device, equipment and computer-readable storage medium. Background Art

[0002] With the rapid development of the Internet, more and more users have changed from simple information recipients to participants in information production. Comments on e-commerce websites record users' evaluations of certain items or logistics services. That is, users can post comments and other descriptive information about an item after purchasing it. Other consumers can browse the description of the item as a reference for decision-making when purchasing the item. Mining and sorting effective descriptions of items allows users to see the most valuable information.

[0003] In the prior art, in order to screen out high-quality and objective description information of items, a logistic regression algorithm is often used to predict the quality of description information. However, the logistic regression algorithm is a linear model and cannot mine the potential semantic connections between a large amount of description information, nor can it accurately sort the description information. Summary of the invention

[0004] The present invention provides a description information processing method, device, equipment and computer-readable storage medium to solve the problem in the prior art that the description quality prediction model cannot mine the potential semantic connections between a large amount of description information and cannot accurately sort the description information.

[0005] One aspect of the present invention is to provide a description information processing method, comprising:

[0006] Get the description information of the target item;

[0007] Preprocessing the description information of the target object to determine the effective description and weight information of the target object;

[0008] Extracting descriptive feature data of a valid description of the target object;

[0009] Inputting the weight information and description feature data of the effective description of the target object into the trained description quality prediction model to calculate the description quality information of the effective description of the target object;

[0010] The valid descriptions of the target object are sorted according to the description quality information of the valid descriptions of the target object.

[0011] Another aspect of the present invention is to provide a description information processing device, comprising:

[0012] A data acquisition module is used to obtain description information of the target object;

[0013] A data preprocessing module, used to preprocess the description information of the target object to determine the effective description of the target object and its weight information;

[0014] A feature extraction module, used to extract descriptive feature data of a valid description of the target object;

[0015] A description quality prediction module, used to input the weight information and description feature data of the effective description of the target object into a trained description quality prediction model, and calculate the description quality information of the effective description of the target object;

[0016] The description information sorting module is used to sort the effective descriptions of the target object according to the description quality information of the effective descriptions of the target object.

[0017] Another aspect of the present invention is to provide a description information processing device, comprising:

[0018] A memory, a processor, and a computer program stored on the memory and executable on the processor;

[0019] Wherein, the processor implements the above-mentioned description information processing method when running the computer program.

[0020] Another aspect of the present invention is to provide a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned description information processing method is implemented.

[0021] The description information processing method, device, equipment and computer-readable storage medium provided by the present invention obtain the description information of the target item; pre-process the description information of the target item to determine the effective description of the target item and its weight information; extract the description feature data of the effective description of the target item; input the weight information and description feature data of the effective description of the target item into a trained description quality prediction model to calculate the description quality information of the effective description of the target item, which can fully explore the effective features of the description information, integrate the description information features of multiple dimensions, accurately predict the description quality information of the effective description of the target item, and then sort the effective description of the target item according to the description quality information of the effective description of the target item, so as to achieve accurate sorting of the effective description of the item. On this basis, the effective description of the target item can be displayed to the user in the order after re-sorting, which can improve the order conversion rate brought by the description information. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A flowchart of a description information processing method provided in Embodiment 1 of the present invention;

[0023] Figure 2 A flow chart of a description information processing method provided in Embodiment 2 of the present invention;

[0024] Figure 3 A schematic diagram of the structure of an information processing device provided in Embodiment 3 of the present invention;

[0025] Figure 4 A schematic diagram of the structure of an information processing device provided in Embodiment 4 of the present invention;

[0026] Figure 5 This is a schematic diagram of the structure of an information processing device provided in Embodiment 5 of the present invention.

[0027] The above drawings have shown clear embodiments of the present invention, which will be described in more detail below. These drawings and text descriptions are not intended to limit the scope of the present invention in any way, but to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0028] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0029] First, the terms involved in the present invention are explained:

[0030] Deep learning: It is a method in machine learning based on data representation learning. It consists of a multi-layer perceptron, which combines the underlying features to form a more abstract high-level representation attribute, a forward calculation and a back-propagation neural network structure to discover the characteristic representation of data distribution.

[0031] ATrank: The full name in English is Attention Rank, which is a deep learning algorithm based on the network attention real-time mechanism. The present invention is used to build a recommendation framework for describing information.

[0032] Recommendation sorting: refers to the e-commerce website mining massive data to create item recall and sorting layers in order to recommend personalized items to users, and recommend items that users are most interested in.

[0033] In addition, the terms "first", "second", etc. are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. In the description of the following embodiments, the meaning of "multiple" is more than two, unless otherwise clearly and specifically defined.

[0034] The following specific embodiments may be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments. The embodiments of the present invention will be described below in conjunction with the accompanying drawings.

[0035] Embodiment 1

[0036] Figure 1 This is a flow chart of a description information processing method provided in Embodiment 1 of the present invention. The embodiment of the present invention provides a description information processing method to solve the problem that the description quality prediction model in the prior art cannot mine the potential semantic connections between a large amount of description information and cannot accurately sort the description information.

[0037] The method in this embodiment is applied to describe an information processing device, which may be a server device of an e-commerce platform, or a mobile terminal such as a smart phone or a tablet computer. In other embodiments, the method may also be applied to other devices. This embodiment is schematically illustrated using a server device of an e-commerce platform as an example.

[0038] like Figure 1 As shown, the specific steps of this method are as follows:

[0039] Step S101: Obtain description information of the target object.

[0040] The target item refers to the item whose description information needs to be sorted, and the target item can be any item.

[0041] For example, the description information of the item may be information such as the user's evaluation of the item, opinions expressed, etc.

[0042] Step S102: pre-process the description information of the target object to determine the effective description of the target object and its weight information.

[0043] The effective description of an item refers to the effective description information retained after data cleaning of the item description information, filtering out invalid descriptions such as duplicate description information, junk description information, description information that does not meet the word count requirements, etc.

[0044] In this embodiment, a uniform initial weight may be set for each description information of an item, and the weight of the description information may be adjusted according to the text content of the description information and the pictures included therein to obtain a re-determined weight of the description information.

[0045] For example, for low-quality descriptive information such as descriptive information whose text content is ancient poetry, descriptive information containing negative sentiment words, and descriptive information whose text content does not match pictures, the weight of these low-quality descriptive information can be reduced; for effective descriptions of descriptive information containing high-frequency Internet terms, their weight can be increased.

[0046] In this step, after determining the effective description of the item, the weight of the effective description may be adjusted according to the text content of the effective description and the pictures included therein to obtain a re-determined weight of the effective description.

[0047] Step S103: extracting descriptive feature data of effective description of the target object.

[0048] In this embodiment, the description feature data at least includes: publisher features, description information quality features, description information image features and description information text features.

[0049] In addition, the descriptive feature data in this embodiment may include descriptive feature data statistically obtained within a preset time period or multiple preset time periods.

[0050] Exemplarily, the description information is an evaluation of an item as an example to illustrate the description feature data. The publisher characteristics include: the number of times the publisher publishes follow-up comments, the number of times the published comments are clicked, the number of replies to other users, the number of users who reply, the number of times experience certificates are received, the number of experience certificates received, the number of times different levels of comments are published, the number of times the published comments are liked, the number of times the published comments are replied, the number of times the published comments are pinned, the average number of words in the published comments, the maximum number of words in the published comments, the average number of words in the published follow-up comments, the maximum number of words in the published follow-up comments, the number of times the published comments are shared, the number of times the video in the published comments is played in full screen, the average number of words in the reply to the question in the published comments, the maximum number of words in the reply to the question in the published comments, the number of times the published comments fail to pass the review, the average value of the time difference between the comment time and the purchase time in the published comments, the maximum value of the time difference between the comment time and the purchase time in the published comments, the minimum value of the time difference between the comment time and the purchase time in the published comments, and so on.

[0051] Features describing information quality include: the number of default positive reviews given by the publisher, the number of spam comments posted by the publisher, the number of comments containing videos posted by the publisher, the number of likes on a comment, the number of users who like the comment, the average number of users who like the comment per day, the number of full-screen plays on the comment, the number of users who play the comment full-screen, the average number of users who play the comment full-screen per day, the number of clicks on a comment, the number of users who click on the comment, the average number of users who click on the comment per day, the number of clicks or expansions on a comment, the number of users who click or expand the comment, the average number of users who click or expand the comment per day, the number of valid reviews on all items, the time difference between the time when the publisher last posted a comment and the time of purchase, the time difference between the time when the publisher last posted a follow-up comment and the time of purchase, etc.

[0052] The image features of the descriptive information include: the number of times the publisher publishes comments containing images, the number of images contained in the comments published by the publisher, the number of images contained in the comments published by the publisher for each item, the number of times the publisher publishes comments containing images for each item, the number of times the image contained in the comment is clicked, the number of users who click on the image contained in the comment, the average number of users who click on the image contained in the comment per day, and so on.

[0053] The description information text features include: the text content of the description information, the number of type tags contained in the description information, the description information word vector, the description information word vector, the description information context features, etc. Among them, the type tags include service, logistics, color, etc., which are used to indicate what the description information is describing; the description information context features can reflect the association relationship of the context in the description information.

[0054] Step S104: input the weight information and description feature data of the effective description of the target object into the trained description quality prediction model to calculate the description quality information of the effective description of the target object.

[0055] In this embodiment, the description quality prediction model is obtained by pre-training a prediction model based on deep learning using a large amount of description information of items as training data. The description quality prediction model is a deep learning model based on a network attention mechanism, for example, the description quality prediction model can be an ATrank model.

[0056] After obtaining the weight information and description feature data of the effective description of the target object, the weight information and description feature data of the effective description of the target object are input into the trained description quality prediction model, and the description quality information of the effective description of the target object is calculated by the description quality prediction model.

[0057] The description quality information may be a comprehensive score of the quality of each description information.

[0058] Step S105: Sort the valid descriptions of the target items according to the description quality information of the valid descriptions of the target items.

[0059] After determining the description quality information of the effective description of the target object, the effective description of the target object is sorted according to the description quality information of the effective description of the target object, so that when the description information of the target object is displayed, the effective description of the target object can be displayed in the order after the reordering.

[0060] The embodiment of the present invention obtains description information of a target item; pre-processes the description information of the target item to determine a valid description of the target item and its weight information; extracts description feature data of the valid description of the target item; inputs the weight information and description feature data of the valid description of the target item into a trained description quality prediction model to calculate the description quality information of the effective description of the target item. The embodiment of the present invention can fully explore the effective features of the description information, integrate the description information features of multiple dimensions, accurately predict the description quality information of the effective description of the target item, and then sort the effective descriptions of the target item according to the description quality information of the effective description of the target item, thereby achieving accurate sorting of the effective descriptions of the items. On this basis, the effective descriptions of the target item can be displayed to the user in the order after re-sorting, thereby improving the order conversion rate brought by the description information.

[0061] Embodiment 2

[0062] Figure 2 A flow chart of the description information processing method provided for the second embodiment of the present invention. Based on the above-mentioned embodiment 1, in this embodiment, the weight information and description feature data of the effective description of the target item are input into the trained description quality prediction model, and before the description quality information of the effective description of the target item is calculated, it also includes: obtaining the description information of all sample items; preprocessing the description information of the sample items to determine the effective description of the sample items and their weight information; extracting the description feature data of the effective description of the sample items; determining the quality label of the effective description of the sample items according to the label indicator information of the effective description of the sample items; training a prediction model based on deep learning according to the weight information, description feature data and quality label of the effective description of the sample items to obtain a description quality prediction model.

[0063] like Figure 2 As shown, the specific steps of this method are as follows:

[0064] Step S201: Obtain description information of all sample items.

[0065] In this embodiment, all or part of the items in the historical data may be used as sample items to obtain description information of the sample items.

[0066] Step S202: pre-process the description information of the sample items to determine the effective description of the sample items and their weight information.

[0067] The process of preprocessing the description information of the sample object is consistent with the process of preprocessing the description information of the target object.

[0068] In this embodiment, the description information of an item is preprocessed to determine a valid description of the item, which can be implemented in the following manner:

[0069] The same description information of the same item in the description information of the item is deduplicated and filtered, and one of the same description information of the same item is retained; junk description information containing sensitive words in the description information of the item is filtered out; description information in the description information of the item whose word count of the description information text is not within a preset word count range is filtered out.

[0070] Optionally, when filtering out duplicates of identical description information of the same item in the description information of the item, the description information with the most recent publishing time may be retained as a valid description in the order of publishing time.

[0071] For example, some users place an order for 10 items at a time, and write "Item received" as a comment for each item. For these repeated and identical comments on the same item, we filter them out and only keep the most recent comment as the valid description of the item.

[0072] Some description information contains sensitive words such as uncivilized language. By setting a sensitive word library, the text content of the description information is matched with the sensitive words in the sensitive word library to determine whether the text content of the description information contains sensitive words; if it is determined that the text content of the description information contains sensitive words, the description information is filtered out.

[0073] The preset word range can be set by technicians according to actual needs and experience, and is not specifically limited in this embodiment. For example, the preset word range can be greater than 5, or the preset word range can be greater than 5 and less than 100, and so on.

[0074] Furthermore, in order to accurately predict the quality of the description information, in this embodiment, the quality weight of the description information may be set, and a uniform initial weight may be set for each piece of description information.

[0075] Preprocess the description information of an item to determine the weight information of the effective description of the item, including: setting an initial weight for the effective description of the item; adjusting the weight of the effective description of the item according to the text content and pictures of the effective description of the item. The higher the weight value of the description information, the higher the quality of the description information, and the higher the description quality information predicted by the description quality prediction model may be, and the higher the ranking of the description information may be in the future.

[0076] Specifically, the weight of the effective description of the item is adjusted according to the text content and the picture of the effective description of the item, including at least one of the following:

[0077] (1) Calculate the similarity between the text content of each valid description of the item and the ancient poetry texts in the preset ancient poetry library; reduce the weight of the valid description whose text content has a similarity with any ancient poetry text in the preset ancient poetry library to reach a first similarity threshold.

[0078] In actual applications, some users do not fill in the description information carefully, but directly input a passage of ancient poetry that they are familiar with as the description information text content. This kind of description information contains little effective information and has low reference value for users.

[0079] In this embodiment, a preset ancient poetry library is maintained to store common ancient poetry. The similarity between the text content of the description information and the ancient poetry text in the preset ancient poetry library is calculated; for the effective description whose text content has a similarity with any ancient poetry text in the preset ancient poetry library that reaches a first similarity threshold, its weight is reduced so that the description information is ranked further back.

[0080] The first similarity threshold can be set by technicians according to actual needs and experience, and is not specifically limited in this embodiment. For example, the first similarity threshold can be 95%.

[0081] In addition, the ancient poems included in the ancient poetry library can be dynamically updated by technical personnel at any time.

[0082] (2) For a valid description containing a picture, a preset text-picture matching model is used to determine whether the text content of the valid description matches the picture; if it is determined that the text content of the valid description does not match the picture, the weight of the valid description is reduced to make the description information ranked lower.

[0083] Exemplarily, the preset text image matching model can adopt a binary classification model, for example, a gradient boosting decision tree (GBDT) model can be adopted. In addition, the preset text image matching model can also be implemented by any model in the prior art for determining whether the text content matches the image, which is not specifically limited in this embodiment.

[0084] (3) Reduce the weight of valid descriptions containing negative sentiment words.

[0085] In this embodiment, a preset negative sentiment vocabulary library is maintained to store common negative sentiment vocabulary.

[0086] For a valid description containing negative sentiment words, the weight of the description information is reduced so that the description information is ranked lower.

[0087] In addition, the negative sentiment vocabulary included in the negative sentiment vocabulary library can be dynamically updated by technical personnel at any time.

[0088] (4) Calculate the similarity between the text content of each valid description of the item and the Internet terms in the preset Internet term library; increase the weight of the valid description whose text content has a similarity with any Internet term in the preset Internet term library that reaches a second similarity threshold.

[0089] In this embodiment, a preset network term library is maintained to store common network terms. The similarity between the text content of the description information and the network terms in the preset network term library is calculated; for the effective description whose text content and the network terms in the preset network term library reach the second similarity threshold, its weight is increased to make the description information be ranked higher.

[0090] The second similarity threshold may be set by technicians according to actual needs and experience, and is not specifically limited in this embodiment.

[0091] In addition, the Internet terms included in the Internet term library can be dynamically updated by technical personnel at any time.

[0092] Step S203: extracting descriptive feature data of effective descriptions of sample items.

[0093] In this embodiment, the description feature data at least includes: publisher features, description information quality features, description information image features and description information text features.

[0094] In addition, the descriptive feature data in this embodiment may include descriptive feature data statistically obtained within a preset time period or multiple preset time periods.

[0095] Exemplarily, the description information is an evaluation of an item as an example to illustrate the description feature data. The publisher characteristics include: the number of times the publisher publishes follow-up comments, the number of times the published comments are clicked, the number of replies to other users, the number of users who reply, the number of times experience vouchers are received, the number of experience vouchers received, the number of times different levels of comments are published, the number of times the published comments are liked, the number of times the published comments are replied, the number of times the published comments are pinned, the average number of words in the published comments, the maximum number of words in the published comments, the average number of words in the published follow-up comments, the maximum number of words in the published follow-up comments, the number of times the published comments are shared, the number of times the video in the published comments is played in full screen, the average number of words in the reply to the question in the published comments, the maximum number of words in the reply to the question in the published comments, the number of times the published comments fail to pass the review, the average value of the time difference between the comment time and the purchase time in the published comments, the maximum value of the time difference between the comment time and the purchase time in the published comments, the minimum value of the time difference between the comment time and the purchase time in the published comments, and so on.

[0096] Features describing information quality include: the number of default positive comments given by the publisher, the number of spam comments posted by the publisher, the number of comments containing videos posted by the publisher, the number of likes on a comment within at least one preset time period, the number of users who liked the comment, the average number of users who liked the comment per day, the number of times the comment was played full screen, the number of users who played the comment full screen, the average number of users who played the comment full screen per day, the number of clicks on a comment, the number of users who clicked on the comment, the average number of users who clicked on the comment per day, the number of clicks or expansions on a comment, the number of users who clicked or expanded the comment, the average number of users who clicked or expanded the comment per day, the number of valid comments among all item reviews, the time difference between the time when the publisher last posted a comment and the time of purchase, the time difference between the time when the publisher last posted a follow-up comment and the time of purchase, etc.

[0097] The image features of the descriptive information include: the number of times the publisher publishes comments containing images, the number of images contained in the comments published by the publisher, the number of images contained in the comments published by the publisher for each item, the number of times the publisher publishes comments containing images for each item, the number of times the image contained in the comment is clicked, the number of users who click on the image contained in the comment, the average number of users who click on the image contained in the comment per day, and so on.

[0098] The descriptive information text features include: the text content of the review, the number of type tags contained in the review, the review word vector, the review word vector, the review context features, etc. Among them, the type tags include service, logistics, color, etc., which are used to describe the content of the review; the review context features can reflect the correlation relationship of the context in the review.

[0099] For example, publisher characteristics may include: the number of times the publisher has posted follow-up reviews in the past 7 days, the past 30 days, and the past 60 days; the number of times the comments posted by the publisher have been clicked in the past 7 days, the past 30 days, and the past 60 days; the number of times the publisher has replied to other users in the past 7 days, the past 30 days, and the past 60 days; the number of users to whom the publisher has replied in the past 7 days, the past 30 days, and the past 60 days; the number of times the publisher has received experience certificates (such as Jingdou) in the past 7 days, the past 30 days, and the past 60 days; the number of experience certificates received by the publisher in the past 7 days, the past 30 days, and the past 60 days (such as the number of Jingdou received); The number of times the publisher posted 5-star reviews; the number of times the publisher posted 4-star reviews in the past 7 days, the past 30 days, and the past 60 days; the number of times the publisher posted 3-star reviews in the past 7 days, the past 30 days, and the past 60 days; the number of times the publisher posted 2-star reviews in the past 7 days, the past 30 days, and the past 60 days; the number of times the publisher posted 1-star reviews in the past 7 days, the past 30 days, and the past 60 days; the number of times the comments posted by the publisher in the past 7 days, the past 30 days, and the past 60 days were liked; the number of times the comments posted by the publisher in the past 7 days, the past 30 days, and the past 60 days were replied to; the number of times the comments posted by the publisher in the past 7 days, the past 30 days, and the past 60 days were pinned; The average number of words in comments posted by publishers in the past 7 days, the past 30 days, and the past 60 days; the maximum number of words in comments posted by publishers in the past 7 days, the past 30 days, and the past 60 days; the average number of words in follow-up comments posted by publishers in the past 7 days, the past 30 days, and the past 60 days; the maximum number of words in follow-up comments posted by publishers in the past 7 days, the past 30 days, and the past 60 days; the number of times comments posted by publishers in the past 7 days, the past 30 days, and the past 60 days have been shared; the number of times videos in comments posted by publishers in the past 7 days, the past 30 days, and the past 60 days have been played in full screen; the number of questions answered by users in comments posted by publishers in the past 7 days, the past 30 days, and the past 60 days The average number of words in questions asked about comments posted by publishers; the maximum number of words in responses to questions about comments posted by publishers in the past 7 days, the past 30 days, and the past 60 days; the number of times comments posted by publishers in the past 7 days, the past 30 days, and the past 60 days failed to pass the review; the average value of the time difference between the comment time and the purchase time in comments posted by publishers in the past 7 days, the past 30 days, and the past 60 days; the maximum value of the time difference between the comment time and the purchase time in comments posted by publishers in the past 7 days, the past 30 days, and the past 60 days; the minimum value of the time difference between the comment time and the purchase time in comments posted by publishers in the past 7 days, the past 30 days, and the past 60 days, etc.

[0100] For example, the description of information quality features may include: the number of default positive comments given by the publisher in the past 7 days, the past 30 days, and the past 60 days; the number of spam comments posted by the publisher in the past 7 days, the past 30 days, and the past 60 days; the number of video comments posted by the publisher in the past 7 days, the past 30 days, and the past 60 days; the number of likes on a comment in the past 7 days, the past 30 days, and the past 60 days; the number of users who liked the comment in the past 7 days, the past 30 days, and the past 60 days; the average number of users who liked the comment in the past 7 days, the past 30 days, and the past 60 days; the number of full-screen playback of the comment in the past 7 days, the past 30 days, and the past 60 days; the number of users who played the comment in full screen in the past 7 days, the past 30 days, and the past 60 days; the average number of comments on the comment in the past 7 days, the past 30 days, and the past 60 days The number of users who played the comment in full screen mode; the number of times the comment was clicked in the past 7 days, the past 30 days, and the past 60 days; the number of users who clicked on the comment in the past 7 days, the past 30 days, and the past 60 days; the average number of users who clicked on the comment per day in the past 7 days, the past 30 days, and the past 60 days; the number of times the comment was clicked or expanded in the past 7 days, the past 30 days, and the past 60 days; the number of users who clicked or expanded the comment in the past 7 days, the past 30 days, and the past 60 days; the average number of users who clicked or expanded the comment per day in the past 7 days, the past 30 days, and the past 60 days; the number of valid comments among the reviews of all items in the past 7 days, the past 30 days, and the past 60 days; the time difference between the time when the publisher last posted a comment and the time of purchase; the time difference between the time when the publisher last posted a follow-up comment and the time of purchase; etc.

[0101] For example, the description information image features may include: the number of times the publisher has posted comments containing pictures in the past 7 days, the past 30 days, and the past 60 days; the number of pictures contained in the comments posted by the publisher in the past 7 days, the past 30 days, and the past 60 days; the number of pictures contained in the comments posted by the publisher for each item in the past 7 days, the past 30 days, and the past 60 days; the number of times the publisher has posted comments containing pictures for each item in the past 7 days, the past 30 days, and the past 60 days; the number of times the pictures contained in the comments have been clicked in the past 7 days, the past 30 days, and the past 60 days; the number of users who clicked on the pictures contained in the comments in the past 7 days, the past 30 days, and the past 60 days; the average number of users who clicked on the pictures contained in the comments in the past 7 days, the past 30 days, and the past 60 days, and the like.

[0102] Step S204: Determine the quality label of the effective description of the sample item according to the label indicator information of the effective description of the sample item.

[0103] In this embodiment, firstly, label index information of the effective description of the sample item is obtained, wherein the label index information includes at least one of the following: the purchase rate, the number of likes, whether the description is pinned to the top, and whether it is set as the best.

[0104] The add-to-cart rate of the description information refers to the probability that the user adds the item corresponding to the description information to the shopping cart after the description information is clicked by the user.

[0105] Exemplarily, the add-to-cart rate of the description information can be calculated as follows: the ratio of the number of times the description information is clicked by the user and the user adds the item corresponding to the description information to the shopping cart to the total number of times the description information is clicked by the user; that is, the add-to-cart rate of the description information = the number of times the description information is clicked by the user and the user adds the item corresponding to the description information to the shopping cart / the total number of times the description information is clicked by the user.

[0106] The number of likes for a description information refers to the number of times the description information has been liked.

[0107] Specifically, numerical label indicator information such as the add-to-cart rate and number of likes of the description information can be directly used as the quality score of the description information.

[0108] Regarding whether the description information is a top description, if the description information is a top description, then the quality score corresponding to the label indicator information of whether the description information is a top description is a first score; if the description information is not a top description, then the quality score corresponding to the label indicator information of whether the description information is a top description is a second score; wherein the first score is greater than the second score, the first score can be 1, and the second score can be 0.

[0109] Regarding whether the description information is set to be the best, if the description information is set to be the best description information, then the quality score corresponding to the label indicator information of whether the description information is set to be the best is the third score; if the description information is not set to be the best description information, then the quality score corresponding to the label indicator information of whether the description information is set to be the best is the fourth score; wherein the third score is greater than the fourth score, the third score can be 1, and the fourth score can be 0.

[0110] In this step, the quality label of the effective description of the sample item is determined according to the label indicator information of the effective description of the sample item, which can be specifically implemented in the following manner:

[0111] The quality score corresponding to each type of label indicator information of each valid description is obtained, and the quality score corresponding to each type of label indicator information is weighted and summed to obtain the quality label of the valid description.

[0112] Optionally, after obtaining the quality score corresponding to each type of label indicator information of the effective description, the quality score corresponding to each type of label indicator information may be directly summed to obtain the quality label of the effective description.

[0113] In addition, other methods may be used to comprehensively and effectively describe the quality score corresponding to each type of label indicator information to obtain an effective quality label, which is not specifically limited in this embodiment.

[0114] Step S205: According to the weight information, description feature data and quality labels of the effective descriptions of the sample items, a prediction model based on deep learning is trained to obtain a description quality prediction model.

[0115] After obtaining the weight information, descriptive feature data and quality labels of the effective description of the sample items, the model parameters of the deep learning-based prediction model are trained using the weight information, descriptive feature data and quality labels of the effective description of the sample items to obtain a description quality prediction model.

[0116] The description quality prediction model in this embodiment is a deep learning model based on a network attention mechanism. Preferably, the description quality prediction model can be an ATrank model.

[0117] The description quality prediction model obtained after training is used to calculate the description quality information of the effective description of the object according to the weight information and description feature data of the effective description of the input object.

[0118] In addition, when training the model parameters of the deep learning-based prediction model, the sample items are divided into a training set, a validation set, and a test set.

[0119] After the description quality prediction model is trained, the effect of sorting description information using quality labels calculated by the trained description quality prediction model can be tested on a test set.

[0120] Specifically, the evaluation indicators of the test include offline accuracy and offline AUC (Area Under Curve).

[0121] The description information on the item details page viewed by the user is ranked and recommended. If the user clicks on the recommended description information and adds the corresponding item to the shopping cart, the recommendation is considered accurate. The accuracy of the description quality prediction model trained in this embodiment on the test set is 0.89, and the offline AUC evaluation index on the test set is 0.91.

[0122] In addition, after the trained description quality prediction model was put into use, manual evaluation and statistical business indicators proved that the order conversion rate was improved after the quality labels calculated by the trained description quality prediction model were used to sort the description information.

[0123] Usually, after obtaining a trained description quality prediction model, you can try it online for a period of time, and count the effect of sorting description information using the quality labels calculated by the trained description quality prediction model during the trial period. If the effect is good, the model will be officially put into use. In addition, the newly launched model and the original model will be tested online at the same time to comprehensively compare various indicators and select the optimal model. Before the trained model is officially launched, the server, client, and model need to be jointly debugged to ensure the correctness of the business logic.

[0124] In addition, after the trained model is put online, technical personnel can promptly feedback abnormal data to relevant technical personnel when they find abnormal situations online, so that relevant technical personnel can analyze the abnormal data to determine whether it is caused by the data, business logic, or model, and make targeted improvements to each type of abnormal situation, and iterate and optimize the model again and again.

[0125] Through the above steps S201-S205, a description quality prediction model can be pre-trained; through steps S206-S210, the description information of the object can be filtered, and the description quality information of the effective description of the object can be calculated by applying the description quality prediction model, and the effective descriptions of the target objects can be sorted according to the description quality information of the effective descriptions of the objects.

[0126] Step S206: Obtain description information of the target item.

[0127] The target item refers to the item whose description information needs to be sorted, and the target item can be any item.

[0128] Step S207: pre-process the description information of the target object to determine the effective description of the target object and its weight information.

[0129] In this embodiment, the process of preprocessing the description information of the target item is consistent with the process of preprocessing the description information of the sample item in the above step S202. The process of preprocessing the description information of an item to determine the effective description and weight information of the item is described in step S202, which will not be repeated here.

[0130] Step S208: extracting descriptive feature data of a valid description of the target object.

[0131] In this embodiment, the description feature data at least includes: publisher features, description information quality features, description information image features and description information text features.

[0132] In this embodiment, the descriptive feature data of the effective description of the target object is consistent with the information type included in the descriptive feature data of the effective description of the sample object in the above step S203, and will not be described in detail here.

[0133] Step S209: input the weight information and description feature data of the effective description of the target object into the trained description quality prediction model to calculate the description quality information of the effective description of the target object.

[0134] In this embodiment, after obtaining the weight information and description feature data of the effective description of the target object, the weight information and description feature data of the effective description of the target object are input into a trained description quality prediction model, and the description quality information of the effective description of the target object is calculated by the description quality prediction model.

[0135] The description quality information may be a comprehensive score of the quality of each description information.

[0136] Step S210: Sort the valid descriptions of the target items according to the description quality information of the valid descriptions of the target items.

[0137] After determining the description quality information of the effective description of the target object, the effective description of the target object is sorted according to the description quality information of the effective description of the target object, so that when the description information of the target object is displayed, the effective description of the target object can be displayed in the order after the reordering.

[0138] The embodiment of the present invention obtains description information of a target item; pre-processes the description information of the target item to determine a valid description of the target item and its weight information; extracts description feature data of the valid description of the target item; inputs the weight information and description feature data of the valid description of the target item into a trained description quality prediction model to calculate the description quality information of the effective description of the target item. The embodiment of the present invention can fully explore the effective features of the description information, integrate the description information features of multiple dimensions, accurately predict the description quality information of the effective description of the target item, and then sort the effective descriptions of the target item according to the description quality information of the effective description of the target item, thereby achieving accurate sorting of the effective descriptions of the items. On this basis, the effective descriptions of the target item can be displayed to the user in the order after re-sorting, thereby improving the order conversion rate brought by the description information.

[0139] Embodiment 3

[0140] Figure 3This is a schematic diagram of the structure of the description information processing device provided in the third embodiment of the present invention. The description information processing device provided in the embodiment of the present invention can execute the processing flow provided in the description information processing method embodiment. Figure 3 As shown, the description information processing device 30 includes: a data acquisition module 301, a data preprocessing module 302, a feature extraction module 303, a description quality prediction module 304 and a description information sorting module 305.

[0141] Specifically, the data acquisition module 301 is used to acquire description information of the target object.

[0142] The data preprocessing module 302 is used to preprocess the description information of the target object to determine the effective description of the target object and its weight information.

[0143] The feature extraction module 303 is used to extract descriptive feature data of effective description of the target object.

[0144] The description quality prediction module 304 is used to input the weight information and description feature data of the effective description of the target object into the trained description quality prediction model to calculate the description quality information of the effective description of the target object.

[0145] The description information sorting module 305 is used to sort the effective descriptions of the target objects according to the description quality information of the effective descriptions of the target objects.

[0146] The device provided in the embodiment of the present invention can be specifically used to execute the method embodiment provided in the above-mentioned embodiment 1, and the specific functions will not be repeated here.

[0147] The embodiment of the present invention obtains description information of a target item; pre-processes the description information of the target item to determine a valid description of the target item and its weight information; extracts description feature data of the valid description of the target item; inputs the weight information and description feature data of the valid description of the target item into a trained description quality prediction model to calculate the description quality information of the effective description of the target item. The embodiment of the present invention can fully explore the effective features of the description information, integrate the description information features of multiple dimensions, accurately predict the description quality information of the effective description of the target item, and then sort the effective descriptions of the target item according to the description quality information of the effective description of the target item, thereby achieving accurate sorting of the effective descriptions of the items. On this basis, the effective descriptions of the target item can be displayed to the user in the order after re-sorting, thereby improving the order conversion rate brought by the description information.

[0148] Embodiment 4

[0149] Figure 4This is a schematic diagram of the structure of an information processing device provided in Embodiment 4 of the present invention. Based on the above Embodiment 3, in this embodiment, Figure 4 As shown, the description information processing device 30 also includes: a model training module 306.

[0150] The model training module 306 is used to:

[0151] Obtain description information of all sample items; preprocess the description information of the sample items to determine the effective description of the sample items and their weight information; extract descriptive feature data of the effective description of the sample items; determine the quality label of the effective description of the sample items based on the label indicator information of the effective description of the sample items; train a prediction model based on deep learning based on the weight information, descriptive feature data and quality label of the effective description of the sample items to obtain a description quality prediction model.

[0152] Optionally, the data preprocessing module is also used to:

[0153] The same description information of the same item in the description information of the item is deduplicated and filtered, and one of the same description information of the same item is retained; junk description information containing sensitive words in the description information of the item is filtered out; description information in the description information of the item whose word count of the description information text is not within a preset word count range is filtered out.

[0154] Optionally, the data preprocessing module is also used to:

[0155] An initial weight is set for the effective description of the item; and the weight of the effective description of the item is adjusted according to the text content and pictures of the effective description of the item.

[0156] Optionally, the data preprocessing module is also used to:

[0157] Calculate the similarity between the text content of each valid description of the item and the ancient poetry text in the preset ancient poetry library; reduce the weight of the valid description whose text content similarity with any ancient poetry text in the preset ancient poetry library reaches a first similarity threshold.

[0158] Optionally, the data preprocessing module is also used to:

[0159] For a valid description containing a picture, a preset text-picture matching model is used to determine whether the text content of the valid description matches the picture; if it is determined that the text content of the valid description does not match the picture, the weight of the valid description is reduced.

[0160] Optionally, the data preprocessing module is also used to:

[0161] Reduce the weight of valid descriptions that contain negative sentiment words.

[0162] Optionally, the data preprocessing module is also used to:

[0163] Calculate the similarity between the text content of each valid description of the item and the network terms in the preset network term library; increase the weight of the valid description whose text content has a similarity with any network term in the preset network term library that reaches a second similarity threshold.

[0164] Optionally, the feature extraction module is also used to:

[0165] Extract publisher features, description information quality features, description information image features and description information text features of effective descriptions of items.

[0166] Optionally, the model training module is also used to:

[0167] Obtain label indicator information of effective descriptions of sample items, where the label indicator information includes at least one of the following: add-to-cart rate, number of likes, whether the description is pinned to the top, and whether it is set as the best.

[0168] Optionally, the description quality prediction model is a deep learning model based on a network attention mechanism.

[0169] The device provided in the embodiment of the present invention can be specifically used to execute the method embodiment provided in the above-mentioned embodiment 2, and the specific functions will not be repeated here.

[0170] The embodiment of the present invention obtains description information of a target item; pre-processes the description information of the target item to determine a valid description of the target item and its weight information; extracts description feature data of the valid description of the target item; inputs the weight information and description feature data of the valid description of the target item into a trained description quality prediction model to calculate the description quality information of the effective description of the target item. The embodiment of the present invention can fully explore the effective features of the description information, integrate the description information features of multiple dimensions, accurately predict the description quality information of the effective description of the target item, and then sort the effective descriptions of the target item according to the description quality information of the effective description of the target item, thereby achieving accurate sorting of the effective descriptions of the items. On this basis, the effective descriptions of the target item can be displayed to the user in the order after re-sorting, thereby improving the order conversion rate brought by the description information.

[0171] Embodiment 5

[0172] Figure 5 This is a schematic diagram of the structure of an information processing device provided in Embodiment 5 of the present invention. Figure 5 As shown, the information processing device 70 includes: a processor 701, a memory 702, and a computer program stored in the memory 702 and executable on the processor 701.

[0173] The processor 701 implements the description information processing method of any of the above method embodiments when running the computer program.

[0174] The embodiment of the present invention obtains description information of a target item; pre-processes the description information of the target item to determine a valid description of the target item and its weight information; extracts description feature data of the valid description of the target item; inputs the weight information and description feature data of the valid description of the target item into a trained description quality prediction model to calculate the description quality information of the effective description of the target item. The embodiment of the present invention can fully explore the effective features of the description information, integrate the description information features of multiple dimensions, accurately predict the description quality information of the effective description of the target item, and then sort the effective descriptions of the target item according to the description quality information of the effective description of the target item, thereby achieving accurate sorting of the effective descriptions of the items. On this basis, the effective descriptions of the target item can be displayed to the user in the order after re-sorting, thereby improving the order conversion rate brought by the description information.

[0175] In addition, an embodiment of the present invention further provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the description information processing method provided by any of the above method embodiments is implemented.

[0176] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0177] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0178] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0179] The above-mentioned integrated unit implemented in the form of a software functional unit can be stored in a computer-readable storage medium. The above-mentioned software functional unit is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to perform some steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0180] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example for illustration. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. The specific working process of the device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0181] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. The description and examples are to be considered exemplary only, and the true scope and spirit of the present invention is indicated by the following claims.

[0182] It should be understood that the present invention is not limited to the exact construction that has been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A description information processing method, characterized in that: include: Get the description information of the target item; Preprocessing the description information of the target object to determine the effective description and weight information of the target object; The weight information is determined based on the similarity between the text content of the effective description and the ancient poetry text in the preset ancient poetry library, whether the text content of the effective description and the pictures contained therein match, whether the text content of the effective description contains negative sentiment words, and the similarity between the text content of the effective description and the network terms in the preset network term library; Extracting descriptive feature data of a valid description of the target object; Inputting the weight information and description feature data of the effective description of the target object into the trained description quality prediction model to calculate the description quality information of the effective description of the target object; sorting the effective descriptions of the target object according to the description quality information of the effective descriptions of the target object; The description quality prediction model is a deep learning model based on the network attention mechanism.

2. The method according to claim 1, characterized in that Before inputting the weight information and description feature data of the effective description of the target object into the trained description quality prediction model to calculate the description quality information of the effective description of the target object, the method further includes: Get the description information of all sample items; Preprocessing the description information of the sample items to determine the effective descriptions of the sample items and their weight information; Extracting descriptive feature data of a valid description of the sample object; Determining a quality label of the effective description of the sample item according to the label indicator information of the effective description of the sample item; According to the weight information, descriptive feature data and quality labels of the effective description of the sample items, a prediction model based on deep learning is trained to obtain a description quality prediction model.

3. The method according to claim 1 or 2, characterized in that: Preprocess the description information of an item to determine the effective description of the item, including: Deleting and filtering the same description information of the same item in the description information of the item, and retaining one of the same description information of the same item; Filter out junk description information containing sensitive words in the description information of the item; Filter out the description information of the item whose description text number is not within the preset word count range.

4. The method according to claim 3, characterized in that: Preprocess the description information of an item to determine the weight information of the effective description of the item, including: Set an initial weight for the item's effective description; Adjust the weight of the effective description of the item based on the text content and pictures of the effective description of the item.

5. The method according to claim 4, characterized in that The step of adjusting the weight of the effective description of the item according to the text content and the picture of the effective description of the item includes: Calculate the similarity between the text content of each valid description of the item and the ancient poetry text in the preset ancient poetry library; The weight of the effective description whose similarity between the text content and any ancient poetry text in the preset ancient poetry library reaches the first similarity threshold is reduced.

6. The method according to claim 4, characterized in that The step of adjusting the weight of the effective description of the item according to the text content and the picture of the effective description of the item includes: For a valid description containing a picture, a preset text-picture matching model is used to determine whether the text content of the valid description matches the picture; If it is determined that the text content of the effective description does not match the image, the weight of the effective description is reduced.

7. The method according to claim 4, characterized in that The step of adjusting the weight of the effective description of the item according to the text content and the picture of the effective description of the item includes: Reduce the weight of valid descriptions that contain negative sentiment words.

8. The method according to claim 4, characterized in that The step of adjusting the weight of the effective description of the item according to the text content and the picture of the effective description of the item includes: Calculate the similarity between the text content of each valid description of the item and the network terms in the preset network term library; The weight of the effective description whose similarity between the text content and any network term in the preset network term library reaches a second similarity threshold is increased.

9. The method according to claim 4, characterized in that Extract descriptive characteristic data for a valid description of the item, including: Extract publisher features, description information quality features, description information image features and description information text features of effective descriptions of items.

10. The method according to claim 2, characterized in that Before determining the quality label of the effective description of the sample item according to the label indicator information of the effective description of the sample item, the method further includes: Obtain label indicator information of the effective description of the sample item, wherein the label indicator information includes at least one of the following: add-to-cart rate, number of likes, whether the description is pinned to the top, and whether it is set as the best.

11. A description information processing device, characterized in that: include: A data acquisition module is used to obtain description information of the target object; A data preprocessing module, used to preprocess the description information of the target item, determine the effective description of the target item and its weight information; the weight information is determined based on the similarity between the text content of the effective description and the ancient poetry text in the preset ancient poetry library, whether the text content of the effective description and the pictures contained therein match, whether the text content of the effective description contains negative emotional words, and the similarity between the text content of the effective description and the network terms in the preset network term library; A feature extraction module, used to extract descriptive feature data of a valid description of the target object; A description quality prediction module, used to input the weight information and description feature data of the effective description of the target object into a trained description quality prediction model, and calculate the description quality information of the effective description of the target object; the description quality prediction model is a deep learning model based on a network attention mechanism; The description information sorting module is used to sort the effective descriptions of the target object according to the description quality information of the effective descriptions of the target object.

12. A description information processing device, characterized in that: include: A memory, a processor, and a computer program stored on the memory and executable on the processor; Wherein, when the processor runs the computer program, the method according to any one of claims 1 to 10 is implemented.

13. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 10 is implemented.

Images