Commodity search method and device

Abstract

The present application shows a product search method and device. Obtain a search text; search for candidate products whose comments are related to the search text among multiple products; classify and screen the candidate products based on the classification model to obtain the target product; recall the target product and display the reason for the recall. The comments on the product are submitted by the majority of users after they purchased and used the product in the historical process, that is, the comments on the product include the subjective use experience of the majority of users on the product in the historical process. Therefore, retrieving the candidate products whose comments are related to the search text and classifying and screening the candidate products to obtain the target product, and displaying the reason for the recall, is more convincing and credible for the current user, and more friendly to the current user, thereby increasing the possibility of the retrieved target product being purchased by the current user, and further improving the user's purchasing experience. Displaying the reason for the recall can increase the user's trust in the retrieval method and can increase the user's stickiness.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a commodity retrieval method and device. Background Art

[0002] Currently, more and more users choose to purchase goods on e-commerce platforms. Users can enter search keywords on the e-commerce platform, and then the e-commerce platform can search for products whose titles include the search keywords from multiple products, and then display the retrieved products for users to choose to purchase.

[0003] However, the inventors have found that product titles are often given by merchants themselves. In this way, product titles contain more descriptions of the merchants, lack sufficient credibility, and are likely to mislead users, resulting in a poor purchasing experience for users. Summary of the invention

[0004] In order to solve the above technical problems, the embodiments of the present application show a product search method and device.

[0005] In a first aspect, an embodiment of the present application shows a product search method, the method comprising:

[0006] Get the search text;

[0007] Retrieving at least one candidate product whose review is related to the search text from among the plurality of products;

[0008] Classify and screen the candidate products based on the classification model to obtain target products;

[0009] The target product is recalled and the reason for the recall of the target product is displayed.

[0010] In an optional implementation, the step of retrieving at least one candidate product whose comments are related to the search text from among the multiple products includes:

[0011] In the search text, identifying a search tag for describing a search subject of the search text;

[0012] At least a candidate product whose review is related to the search tag is searched for among a plurality of products.

[0013] In an optional implementation, classifying and screening the candidate commodities based on the classification model to obtain the target commodities includes:

[0014] Determine the quality reviews for each candidate product based on the classification model;

[0015] At least one target product whose high-quality reviews are related to the search tag is searched among the candidate products.

[0016] In an optional implementation, determining the high-quality reviews of each candidate product based on the classification model includes:

[0017] For each candidate product, comments from high-quality users are selected from the comments on the candidate product based on the classification model, and high-quality comments on the candidate product are determined based on the comments from the high-quality users based on the classification model.

[0018] In an optional implementation, selecting the comments of high-quality users from the comments of the candidate products based on the classification model includes:

[0019] Determining high-quality users among users who comment on the candidate product based on the classification model;

[0020] Obtain the comments of the high-quality users on the candidate products.

[0021] In an optional implementation, determining high-quality users among users who comment on the candidate products based on the classification model includes:

[0022] For each user who comments on the candidate product, the user's historical operational reputation, account points and historical interactivity are obtained; based on the classification model, the user's quality score is obtained according to the historical operational reputation, account points and historical interactivity; if the user's quality score is greater than a first preset threshold, the user is determined to be a high-quality user.

[0023] In an optional implementation, determining the high-quality reviews of the candidate product based on the reviews of high-quality users based on the classification model includes:

[0024] Select reviews from high-quality users that include at least one of product information, logistics information, and merchant information;

[0025] Determine the high-quality reviews of the candidate product based on the selected reviews based on the classification model.

[0026] In an optional implementation, determining the high-quality reviews of the candidate product based on the selected reviews based on the classification model includes:

[0027] Obtain the semantic fluency of each selected comment based on the classification model;

[0028] Among the selected comments, screening comments whose semantic fluency is greater than or equal to a second preset threshold;

[0029] Determine high-quality reviews for the candidate product based on the filtered reviews.

[0030] In an optional implementation, obtaining the semantic coherence of each selected comment based on the classification model includes:

[0031] Use the language model LM to obtain the semantic fluency of each comment.

[0032] In an optional implementation, the method further includes:

[0033] A sample comment set is obtained, wherein the sample comment set includes at least one sample comment annotated with semantic coherence; and a model is trained using the sample comments in the sample comment set until all weights in the model converge to obtain the language model.

[0034] In an optional implementation, determining the high-quality reviews of the candidate product based on the selected reviews based on the classification model includes:

[0035] Obtaining the relevance between each selected review and the candidate product based on the classification model;

[0036] Among the selected reviews, screening reviews whose relevance to the candidate product is greater than or equal to a third preset threshold;

[0037] Determine high-quality reviews for the candidate product based on the filtered reviews.

[0038] In an optional implementation, obtaining the relevance between each selected review and the candidate product based on the classification model includes:

[0039] Obtain a topic for describing the industry to which the candidate product belongs;

[0040] For each review, the relevance between the review and the theme of the industry is obtained based on the text topic model, and is used as the relevance between the review and the candidate product.

[0041] In an optional implementation, determining the high-quality reviews of the candidate product based on the selected reviews based on the classification model includes:

[0042] Obtaining the semantic fluency of each selected comment based on the classification model, and obtaining the relevance between each selected comment and the candidate product based on the classification model;

[0043] Among the selected reviews, screening reviews whose semantic coherence is greater than or equal to a second preset threshold and whose relevance to the candidate product is greater than or equal to a third preset threshold;

[0044] Determine high-quality reviews for the candidate product based on the filtered reviews.

[0045] In an optional implementation, searching at least one target product among candidate products for a product with high-quality reviews related to the search tag includes:

[0046] For each candidate product, identifying a review tag for describing a review topic of the high-quality review in the high-quality reviews of the candidate product;

[0047] If the review tag includes the search tag, the candidate product is determined as a target product whose high-quality reviews are related to the search tag.

[0048] In an optional implementation, the method further includes:

[0049] The comment labels of the candidate commodities and the candidate commodities are formed into corresponding table items, and stored in the corresponding relationship between the comment labels of the commodity reviews and the commodities.

[0050] In an optional implementation, searching at least one target product among candidate products for a product with high-quality reviews related to the search tag includes:

[0051] In the correspondence between the comment tags of the product reviews and the product, searching for the comment tags related to the search tag;

[0052] A candidate product corresponding to the found review tag is searched in the corresponding relationship and used as the target product.

[0053] In an optional implementation, identifying, in the search text, a search tag for describing a search subject of the search text includes:

[0054] In the search text, a search tag for describing a search subject of the search text is identified based on a search tag identification model.

[0055] In an optional implementation, the method further includes:

[0056] Acquire a sample search text set, wherein the sample search text set includes at least one sample search text annotated with a sample search tag;

[0057] The model is trained using the sample retrieval texts in the sample retrieval text set until all weights in the model converge, thereby obtaining the retrieval tag recognition model.

[0058] In an optional implementation, the displaying of the reason for recalling the target product includes:

[0059] Among the reviews of the target product, counting the number of reviews related to the search tag;

[0060] Generate the recall reason according to the search tag and the number of comments;

[0061] Present the reason for the recall.

[0062] In an optional implementation, generating the recall reason according to the search tag and the number of comments includes:

[0063] Obtaining a recall reason generation template, wherein the recall reason generation template includes a comment quantity filling space and a search tag filling space;

[0064] The number of comments is filled in the comment number filling space in the recall reason generation template, and the search tag is filled in the search tag filling space, so as to obtain the recall reason.

[0065] In an optional implementation, the displaying of the reason for recalling the target product includes:

[0066] Obtaining reviews of the target product;

[0067] generating the recall reason according to the review of the target product;

[0068] Present the reason for the recall.

[0069] In an optional implementation, obtaining the review of the target product includes:

[0070] Among the reviews of the target product, the reviews related to the search tag are counted.

[0071] In an optional implementation, generating the recall reason according to the comment includes:

[0072] The recall reason generation model is used to generate the recall reason according to the review.

[0073] In an optional implementation, the method further includes:

[0074] Obtain multiple sample review sets, each sample review set including multiple sample reviews annotated with the same sample recall reason;

[0075] The model is trained using sample comments in multiple sample comment sets until all weights in the model converge, thereby obtaining the recall reason generation model.

[0076] In a second aspect, an embodiment of the present application shows a commodity search device, including:

[0077] The acquisition module is used to obtain the search text;

[0078] A search module, used to search at least one candidate product whose comments are related to the search text from among a plurality of products;

[0079] A screening module, used to classify and screen the candidate commodities based on a classification model to obtain target commodities;

[0080] A recall module, used for recalling the target product;

[0081] The display module is used to display the reason for recalling the target product.

[0082] In an optional implementation, the retrieval module includes:

[0083] An identification submodule, used to identify, in the search text, a search tag used to describe a search subject of the search text;

[0084] The search submodule is used to search at least one candidate product whose comments are related to the search tag from among multiple products.

[0085] In an optional implementation, the screening module includes:

[0086] A determination submodule is used to determine the quality reviews of each candidate product based on the classification model;

[0087] The search submodule is used to search at least one target product whose high-quality reviews are related to the search tag among the candidate products.

[0088] In an optional implementation, the determining submodule includes:

[0089] The selection unit is used to select, for each candidate product, comments from high-quality users among the comments of the candidate product based on the classification model; the first determination unit is used to determine the high-quality comments of the candidate product according to the comments of the high-quality users based on the classification model.

[0090] In an optional implementation, the selection unit includes:

[0091] A first determination subunit, configured to determine high-quality users among users who comment on the candidate product based on a classification model;

[0092] The acquisition subunit is used to acquire the comments of the high-quality users on the candidate products.

[0093] In an optional implementation, the first determination subunit is specifically used to: for each user who comments on the candidate product, obtain the user's historical operation reputation, account points and historical interaction; obtain the user's quality score based on the historical operation reputation, account points and historical interaction based on a classification model; if the user's quality score is greater than a first preset threshold, determine the user as a high-quality user.

[0094] In an optional implementation, the determining unit includes:

[0095] A selection subunit, configured to select, from among the reviews of high-quality users, reviews that include at least one of product information, logistics information, and merchant information;

[0096] The second determining subunit is used to determine the high-quality reviews of the candidate product according to the selected reviews based on the classification model.

[0097] In an optional implementation, the second determination subunit is specifically used to: obtain the semantic coherence of each selected review based on the classification model; filter out reviews whose semantic coherence is greater than or equal to a second preset threshold among the selected reviews; and determine the high-quality reviews of the candidate product based on the filtered reviews.

[0098] In an optional implementation, the second determining subunit is specifically configured to: use a language model LM to obtain the semantic fluency of each comment respectively.

[0099] In an optional implementation, the second determination subunit is specifically used to: obtain a sample comment set, the sample comment set including at least one sample comment annotated with semantic coherence; use the sample comments in the sample comment set to train the model until the weights in the model converge to obtain the language model.

[0100] In an optional implementation, the second determination subunit is specifically used to: obtain the correlation between each selected review and the candidate product based on the classification model; among the selected reviews, filter out reviews whose correlation with the candidate product is greater than or equal to a third preset threshold; and determine the high-quality reviews of the candidate product based on the filtered reviews.

[0101] In an optional implementation, the second determination subunit is specifically used to: obtain a theme for describing the industry in which the candidate product is located; for each comment, obtain the correlation between the comment and the theme of the industry based on a text topic model, and use it as the correlation between the comment and the candidate product.

[0102] In an optional implementation, the second determination subunit is specifically used to: obtain the semantic coherence of each selected comment based on the classification model, and obtain the relevance between each selected comment and the candidate product based on the classification model; among the selected comments, filter out comments whose semantic coherence is greater than or equal to a second preset threshold and whose relevance to the candidate product is greater than or equal to a third preset threshold; determine the high-quality comments of the candidate product based on the filtered comments.

[0103] In an optional implementation, the search submodule includes:

[0104] A first identification unit is used to identify, for each candidate product, a review tag used to describe a review topic of the high-quality review in the high-quality reviews of the candidate product;

[0105] The second determining unit is configured to determine the candidate product as a target product whose high-quality reviews are related to the search tag if the review tag includes the search tag.

[0106] In an optional implementation, the search submodule further includes:

[0107] The storage unit is used to form a corresponding table item with the comment label of the candidate product and the candidate product, and store it in the corresponding relationship between the comment label of the product review and the product.

[0108] In an optional implementation, the search submodule includes:

[0109] A first search unit, configured to search for a comment tag related to the search tag in the correspondence between the comment tags of the comments of the product and the product;

[0110] The second searching unit is used to search for a candidate commodity corresponding to the found review tag in the corresponding relationship and use it as the target commodity.

[0111] In an optional implementation, the identification submodule includes:

[0112] The second recognition unit is used to recognize, in the search text, a search tag for describing a search topic of the search text based on a search tag recognition model.

[0113] In an optional implementation, the identified submodule further includes:

[0114] A first acquisition unit, configured to acquire a sample search text set, wherein the sample search text set includes at least one sample search text annotated with a sample search tag;

[0115] The first training unit is used to train the model using the sample retrieval texts in the sample retrieval text set until all weights in the model converge to obtain the retrieval tag recognition model.

[0116] In an optional implementation, the display module includes:

[0117] A statistics submodule, used for counting the number of comments related to the search tag in the comments of the target product;

[0118] A first generating submodule, used for generating the recall reason according to the search tag and the number of comments;

[0119] The first display submodule is used to display the reason for the recall.

[0120] In an optional implementation, the generating submodule includes:

[0121] A second acquisition unit is used to acquire a recall reason generation template, wherein the recall reason generation template includes a comment quantity filling space and a search tag filling space;

[0122] The filling unit is used to fill in the number of comments in the comment number filling position in the recall reason generation template, and fill in the search tag in the search tag filling position, so as to obtain the recall reason.

[0123] In an optional implementation, the display module includes:

[0124] An acquisition submodule, used to acquire comments on the target product;

[0125] A second generating submodule, configured to generate the recall reason according to the comments of the target product;

[0126] The second display submodule is used to display the reason for the recall.

[0127] In an optional implementation, the acquisition submodule is specifically used to: count the comments related to the search tag in the comments of the target product.

[0128] In an optional implementation, the second generating submodule includes:

[0129] A generating unit is used to generate the recall reason according to the comment using the recall reason generating model.

[0130] In an optional implementation, the second generating submodule further includes:

[0131] A third acquisition unit is used to acquire multiple sample review sets, each sample review set including multiple sample reviews marked with the same sample recall reason;

[0132] The second training unit is used to train the model using sample comments in multiple sample comment sets until all weights in the model converge to obtain the recall reason generation model.

[0133] In a third aspect, an embodiment of the present application shows an electronic device, the electronic device comprising:

[0134] Processor; and

[0135] A memory having executable codes stored thereon, which, when executed, causes the processor to execute the commodity retrieval method as described in the first aspect.

[0136] In a fourth aspect, an embodiment of the present application illustrates one or more machine-readable media on which executable codes are stored. When the executable codes are executed, the processor executes the commodity retrieval method as described in the first aspect.

[0137] Compared with the prior art, the embodiments of the present application have the following advantages:

[0138] In the present application, a search text is obtained; at least candidate products whose comments are related to the search text are searched from multiple products; the candidate products are classified and screened based on the classification model to obtain the target product; the target product is recalled and the reason for recalling the target product is displayed. Among them, the comments on the product are submitted to the electronic device by the majority of users in the historical process after purchasing the product in the electronic device and using the product, that is, the comments on the product include the subjective use experience of the majority of users on the product in the historical process. Therefore, the candidate products whose comments are related to the search text are retrieved and the candidate products are classified and screened to obtain the target product, and the reason for recalling the target product is displayed. This is more convincing and credible for the current user, and more friendly to the current user, thereby increasing the possibility of the retrieved target product being purchased by the current user, and further improving the user's purchasing experience. Furthermore, displaying the reason for recall can increase the user's trust in the search method and increase the user's stickiness. BRIEF DESCRIPTION OF THE DRAWINGS

[0139] Figure 1 The figure is a flow chart of a commodity search method according to an exemplary embodiment.

[0140] Figure 2 The figure is a flow chart of a commodity search method according to an exemplary embodiment.

[0141] Figure 3The figure is a flow chart of a commodity search method according to an exemplary embodiment.

[0142] Figure 4 The figure is a flow chart of a commodity search method according to an exemplary embodiment.

[0143] Figure 5 The figure is a flow chart of a commodity search method according to an exemplary embodiment.

[0144] Figure 6 The figure is a block diagram of a commodity search device according to an exemplary embodiment.

[0145] Figure 7 The figure is a block diagram of a commodity search device according to an exemplary embodiment. DETAILED DESCRIPTION

[0146] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0147] Figure 1 is a flowchart of a commodity search method according to an exemplary embodiment. Figure 1 As shown, the method is used in an electronic device, which includes a front-end server that can be directly operated by a user and a back-end server that cannot be directly operated by the user. The method includes the following steps.

[0148] In step S101, a search text is obtained;

[0149] In the present application, when a user needs to search for a product on the Internet, he or she may submit a search text for searching the product to the electronic device.

[0150] In one case, the user can open a browser in the front-end server, enter a product search page in the browser, and then enter a search text for searching the product in an input box in the product search page. The front-end server then receives the search text entered by the user in the input box in the product search page.

[0151] For example, when a user needs to buy a sweater, if the user attaches great importance to the feel of the sweater, the user can enter the search text "sweater with good feel" in the input box on the product search page. The front-end server receives the search text "sweater with good feel" entered by the user in the input box on the product search page.

[0152] In another case, the user can enter the product search page in the browser of the terminal, and then enter the search text for searching the product in the input box in the product search page. Then the terminal receives the search text entered by the user in the input box in the product search page, and sends the search text to the backend server, which then receives the search text.

[0153] For example, when a user needs to buy a sweater, if the user attaches great importance to the feel of the sweater, the user can enter the search text "sweater with good feel" in the input box on the product search page. The terminal receives the search text "sweater with good feel" entered by the user in the input box on the product search page, and sends the search text "sweater with good feel" to the background server. The background server receives the search text "sweater with good feel".

[0154] In step S102, at least candidate products whose comments are related to the search text are retrieved from among the multiple products;

[0155] In the present application, candidate products whose comments include the search text may be retrieved from multiple products, or candidate products whose comments express the same or similar semantics as the search text may be retrieved.

[0156] In the present application, multiple commodities are stored in the electronic device. In the historical process, every time a user purchases a commodity in the electronic device, the user can comment on the commodity. If the user comments on the commodity, the electronic device will form a corresponding table item with the user's comment on the commodity and store it in the corresponding relationship between the comment and the commodity.

[0157] Thus, in the present application, the electronic device can retrieve candidate products whose comments are related to the search text among multiple products. For example, it can search for comments related to the search text in the correspondence between comments and products, and then search for candidate products corresponding to the comments related to the search text in the correspondence between comments and products.

[0158] Of course, in addition to searching for candidate products whose reviews are related to the search text among a plurality of products, it is also possible to search for candidate products whose titles, etc. are related to the search text among a plurality of products.

[0159] In step S103, the candidate commodities are classified and screened based on the classification model to obtain the target commodity;

[0160] This step may be specifically described in the following embodiments and will not be described in detail here.

[0161] In step S104, the target product is recalled and the reason for recalling the target product is displayed.

[0162] In one embodiment of the present application, if the electronic device is a front-end server, the front-end server can display the target product and the reason for the recall of the target product on the screen of the front-end server, and the user can view the target product and the reason for the recall of the target product on the screen of the front-end server.

[0163] In another embodiment of the present application, if the electronic device is a background server, the background server can send the target product and the reason for the recall of the target product to the terminal used by the user, so that the terminal can display the target product and the reason for the recall of the target product on the screen of the terminal, and the user can view the target product and the reason for the recall of the target product on the screen of the terminal.

[0164] In the present application, a search text is obtained; at least candidate products whose comments are related to the search text are searched from multiple products; the candidate products are classified and screened based on the classification model to obtain the target product; the target product is recalled and the reason for recalling the target product is displayed. Among them, the comments on the product are submitted to the electronic device by the majority of users in the historical process after purchasing the product in the electronic device and using the product, that is, the comments on the product include the subjective use experience of the majority of users on the product in the historical process. Therefore, the candidate products whose comments are related to the search text are retrieved and the candidate products are classified and screened to obtain the target product, and the reason for recalling the target product is displayed. This is more convincing and credible for the current user, and more friendly to the current user, thereby increasing the possibility of the retrieved target product being purchased by the current user, and further improving the user's purchasing experience. Furthermore, displaying the reason for recall can increase the user's trust in the search method and increase the user's stickiness.

[0165] In another embodiment of the present application, see Figure 2 , step S102 includes:

[0166] In step S201, in the search text, a search tag for describing the search subject of the search text is identified;

[0167] In the present application, the retrieval label recognition model can be trained in advance, and the specific training process includes: obtaining a sample retrieval text set, the sample retrieval text set includes at least one sample retrieval text marked with a sample retrieval label; using the sample retrieval text in the sample retrieval text set to train the model until the weights in the model converge, thereby obtaining a retrieval label recognition model.

[0168] Thus, in the present application, the search tag for describing the search subject of the search text can be identified in the search text based on the search tag identification model. For example, the search text is input into the search tag identification model to obtain the search tag for describing the search subject of the search text output by the search tag identification model.

[0169] In step S202, at least candidate products whose reviews are related to the search tag are searched for from among a plurality of products.

[0170] In the present application, among multiple commodities, commodities whose comments include the search tag can be searched, or commodities whose comments express the same or similar semantics as the search tag can be searched.

[0171] In another embodiment of the present application, see Figure 3 , step S103 includes:

[0172] In step S301, high-quality reviews of each candidate product are determined based on the classification model;

[0173] For any candidate product, the quality reviews of the candidate product can be determined in the following way, and the same is true for each other candidate product. The specific process includes:

[0174] 3011. Select comments from high-quality users from the comments on the candidate product based on the classification model.

[0175] When selecting the comments of high-quality users, the high-quality users can be determined from the users who have commented on the candidate product based on the classification model, and then the comments of the high-quality users on the candidate product can be obtained.

[0176] Among them, when determining high-quality users among users who have commented on the candidate product, for any user who has commented on the candidate product, the user's historical operation reputation, account points and historical interaction can be obtained, and then the user's quality score is obtained based on the classification model according to the historical operation reputation, account points and historical interaction; if the user's quality score is greater than a first preset threshold, the user is determined to be a high-quality user, and the same is true for every other user who has commented on the candidate product.

[0177] Among them, when obtaining the quality score of the user, the historical operation reputation, account points and historical interaction can be weightedly summed to obtain the quality score of the user, or the historical operation reputation, account points and historical interaction can be polynomially summed to obtain the quality score of the user, or the quality score of the user can be obtained based on the historical operation reputation, account points and historical interaction based on the LR (Logistic Regressive) model. The present application does not limit the specific acquisition method.

[0178] Among them, the historical operation reputation value includes the shopping reputation value, etc. Whenever a user purchases a product on an electronic device in the historical process, the electronic device will evaluate the user's behavior in the entire process of purchasing the product, and then update the user's shopping reputation value based on the evaluation result.

[0179] In the historical process, every time a user performs an operation on an electronic device, the electronic device will update the user's account points based on the operation. For example, when a user purchases a product on an electronic device, the electronic device will increase the user's account points. When a user comments on the purchased product on an electronic device, the electronic device will increase the user's account points. When a user insults the seller of the product, the electronic device will reduce the user's account points, etc.

[0180] Users can interact and communicate with sellers on electronic devices, interact and communicate with the customer service of electronic devices, and share products or sellers with other users. These interactive behaviors can increase user interactivity.

[0181] 3012. Determine the high-quality reviews of the candidate product based on the reviews of high-quality users based on the classification model.

[0182] In one embodiment, the comments of high-quality users may be used as high-quality comments of the candidate product.

[0183] In another embodiment, after a user purchases a product on an electronic device, a courier company is usually required to pick up the product from the seller and deliver it to the user. That is, in a purchase process, information such as the product, the seller of the product, and the logistics of the courier company are usually involved. Therefore, when a user comments on a purchase, comments involving at least one of the product, the seller of the product, and the logistics of the courier company are valid for other users. Therefore, comments including at least one of the product information, logistics information, and merchant information can be selected from the comments of high-quality users; based on the classification model, the high-quality comments of the candidate product are determined according to the selected comments. For example, in one example, the selected comments can be used as high-quality comments for the product.

[0184] In another example, when a user comments on a product after purchasing it on an electronic device, the comment he input may not be fluent, or may not conform to normal grammatical habits, resulting in low semantic fluency of the product. Other users may not be able to quickly understand the meaning of the comment when seeing the comment. Therefore, such comment has low utility for other users. Therefore, the semantic fluency of each selected comment can be obtained based on the classification model; among the selected comments, the comments with a semantic fluency greater than or equal to a second preset threshold are screened; and the high-quality comments of the candidate product are determined based on the screened comments, for example, the screened comments are used as the high-quality comments of the candidate product.

[0185] The language model can be trained in advance. The training method of the language model includes: obtaining a sample comment set, wherein the sample comment set includes at least one sample comment annotated with semantic fluency; using the sample comments in the sample comment set to train the model until the weights in the model converge to obtain a language model. In this way, the language model can be used to obtain the semantic fluency of each comment. The model includes LM (Language Model) and the like.

[0186] In another example, when some users make a comment on a certain product, the comments they input may sometimes be unrelated to the product, for example, they may be some advertisements for other aspects or descriptions of other products, etc. Such comments are of no use to other users in purchasing the product. Therefore, the relevance between each selected comment and the candidate product can be obtained based on the classification model; then, among the selected comments, the comments whose relevance to the candidate product is greater than or equal to a third preset threshold are screened; and the high-quality comments for the candidate product are determined based on the screened comments, for example, the screened comments are used as the high-quality comments for the candidate product.

[0187] Among them, a theme for describing the industry that the candidate product belongs to may be obtained. For example, when listing products in the electronic device in advance, the staff of the electronic device or the seller of the product may set a theme for describing the industry that the candidate product belongs to, such as medical, food, and office stationery.

[0188] For any comment, the relevance between the comment and the industry theme can be obtained from the text topic model and used as the relevance between the comment and the candidate product. The same is true for each other comment, and the specific form of the text topic model is not limited in this application.

[0189] In another example, the semantic coherence of each selected comment can be obtained based on the classification model, and the relevance between each selected comment and the candidate product can be obtained based on the classification model; among the selected comments, the comments whose semantic coherence is greater than or equal to the second preset threshold and whose relevance to the candidate product is greater than or equal to the third preset threshold are screened; and the high-quality comments of the candidate product are determined based on the screened comments. For example, the screened comments are used as the high-quality comments of the candidate product.

[0190] In step S302, at least a target product whose high-quality reviews are related to the search tag is searched among candidate products.

[0191] In the present application, target products whose quality reviews include the search tag can be retrieved from candidate products, or target products whose quality reviews express semantics that are identical or similar to the semantics expressed by the search tag can be retrieved.

[0192] For any candidate product, a review tag describing the review topic of the high-quality review is identified in the high-quality review of the candidate product; if the review tag includes the search tag, the candidate product is determined as a target product whose high-quality review is related to the search tag. The same is true for each other candidate product.

[0193] For example, suppose a review of a candidate product is "This dress is of the right size and feels good", and the review focuses on describing that the dress is of the right size and feels good. Therefore, the review tags used to describe the review topic of the review include "of the right size" and "feels good".

[0194] Among them, a neural network model is usually used to identify comment tags for describing comment topics of high-quality reviews. However, the process of using the network model to identify comment tags for describing comment topics of high-quality reviews in the high-quality reviews of the product takes a long time. If the network model is used in real time to identify comment tags for describing comment topics of high-quality reviews every time a product is retrieved for a user, it will take a long time, resulting in low retrieval efficiency.

[0195] Therefore, in order to improve the retrieval efficiency when searching for products for users in the future, in another embodiment of the present application, the comment tag of the candidate product can be composed of a corresponding table item with the candidate product, and stored in the correspondence between the comment tag of the product review and the product.

[0196] In this way, when searching for products for users later, if it is necessary to identify the comment tags used to describe the review topics of the high-quality reviews of the product, the comment tags related to the search tag can be directly found in the correspondence between the comment tags of the product reviews and the product, and then the candidate product corresponding to the found comment tags can be found in the correspondence and used as the target product. In this way, the comment tags used to describe the review topics of the high-quality reviews of the candidate products can be identified without using the network model, thereby improving the retrieval efficiency.

[0197] In another embodiment of the present application, see Figure 4 , display the reasons for the recall of the target product, including:

[0198] In step S401, among the reviews of the target product, the number of reviews related to the search tag is counted;

[0199] In the present application, the number of comments including the search tag can be counted among the comments of the target product, or the number of comments whose semantics are the same or similar to those expressed by the search tag can be counted.

[0200] In step S402, a recall reason is generated according to the search tag and the number of comments;

[0201] A recall reason generation template can be obtained, which includes a comment quantity filling space and a search tag filling space; the number of comments is filled in the comment quantity filling space in the recall reason generation template, and the search tag is filled in the search tag filling space, so as to obtain the recall reason.

[0202] For example, a recall reason generation template includes "AAA users think that the product is BBB", where "AAA" is the space for filling in the number of comments and "BBB" is the space for filling in the search tag. Suppose the search tag is "good feel", and among the comments on the target product, the number of comments including the search tag "good feel" is 21; then fill the search tag "good feel" and the number of comments "21" into the recall reason generation template "AAA users think that the product is BBB", and obtain the recall reason "21 users think that the product feels good".

[0203] In step S403, the reason for the recall is displayed.

[0204] In another embodiment of the present application, see Figure 5 , display the reasons for the recall of the target product, including:

[0205] In step S501, the comments of the target product are obtained;

[0206] In one embodiment, in order to improve the accuracy of the generated recall reasons, it is also possible to count the comments related to the search tag in the comments of the target product, or to count the comments whose semantics are the same or similar to the semantics expressed by the search tag.

[0207] In step S502, a recall reason is generated based on the reviews of the target product;

[0208] In the present application, the recall reason generation model can be trained in advance, and the specific training method includes: obtaining multiple sample comment sets, each sample comment set includes multiple sample comments annotated with the same sample recall reason; using the sample comments in the multiple sample comment sets to train the model until the weights in the model converge, and obtain the recall reason generation model. Among them, the model includes a model based on RNN (Recurrent Neural Network, recurrent neural network), etc.

[0209] Thus, in this step, the recall reason generation model can be used to generate the recall reason based on the review. For example, the review of the target product is input into the recall reason generation model to obtain the recall reason output by the recall reason generation model.

[0210] In step S503, the reason for the recall is displayed.

[0211] Figure 6 is a block diagram of a commodity search device according to an exemplary embodiment. Figure 6 As shown, the device comprises:

[0212] An acquisition module 11 is used to acquire a search text;

[0213] A search module 12, configured to search at least one candidate product whose comments are related to the search text from among a plurality of products;

[0214] A screening module 13 is used to classify and screen the candidate commodities based on a classification model to obtain target commodities;

[0215] A recall module 14, used to recall the target product;

[0216] The display module 15 is used to display the reason for recalling the target product.

[0217] In an optional implementation, the retrieval module includes:

[0218] An identification submodule, used to identify, in the search text, a search tag used to describe a search subject of the search text;

[0219] The search submodule is used to search at least one candidate product whose comments are related to the search tag from among multiple products.

[0220] In an optional implementation, the screening module includes:

[0221] A determination submodule is used to determine the quality reviews of each candidate product based on the classification model;

[0222] The search submodule is used to search at least one target product whose high-quality reviews are related to the search tag among the candidate products.

[0223] In an optional implementation, the determining submodule includes:

[0224] The selection unit is used to select, for each candidate product, comments from high-quality users among the comments of the candidate product based on the classification model; the first determination unit is used to determine the high-quality comments of the candidate product according to the comments of the high-quality users based on the classification model.

[0225] In an optional implementation, the selection unit includes:

[0226] A first determination subunit, configured to determine high-quality users among users who comment on the candidate product based on a classification model;

[0227] The acquisition subunit is used to acquire the comments of the high-quality users on the candidate products.

[0228] In an optional implementation, the first determination subunit is specifically used to: for each user who comments on the candidate product, obtain the user's historical operation reputation, account points and historical interaction; obtain the user's quality score based on the historical operation reputation, account points and historical interaction based on a classification model; if the user's quality score is greater than a first preset threshold, determine the user as a high-quality user.

[0229] In an optional implementation, the determining unit includes:

[0230] A selection subunit, configured to select, from among the reviews of high-quality users, reviews that include at least one of product information, logistics information, and merchant information;

[0231] The second determining subunit is used to determine the high-quality reviews of the candidate product according to the selected reviews based on the classification model.

[0232] In an optional implementation, the second determination subunit is specifically used to: obtain the semantic coherence of each selected review based on the classification model; filter out reviews whose semantic coherence is greater than or equal to a second preset threshold among the selected reviews; and determine the high-quality reviews of the candidate product based on the filtered reviews.

[0233] In an optional implementation, the second determining subunit is specifically configured to: use a language model LM to obtain the semantic fluency of each comment respectively.

[0234] In an optional implementation, the second determination subunit is specifically used to: obtain a sample comment set, the sample comment set including at least one sample comment annotated with semantic coherence; use the sample comments in the sample comment set to train the model until the weights in the model converge to obtain the language model.

[0235] In an optional implementation, the second determination subunit is specifically used to: obtain the correlation between each selected review and the candidate product based on the classification model; among the selected reviews, filter out reviews whose correlation with the candidate product is greater than or equal to a third preset threshold; and determine the high-quality reviews of the candidate product based on the filtered reviews.

[0236] In an optional implementation, the second determination subunit is specifically used to: obtain a theme for describing the industry in which the candidate product is located; for each comment, obtain the correlation between the comment and the theme of the industry based on a text topic model, and use it as the correlation between the comment and the candidate product.

[0237] In an optional implementation, the second determination subunit is specifically used to: obtain the semantic coherence of each selected comment based on the classification model, and obtain the relevance between each selected comment and the candidate product based on the classification model; among the selected comments, filter out comments whose semantic coherence is greater than or equal to a second preset threshold and whose relevance to the candidate product is greater than or equal to a third preset threshold; determine the high-quality comments of the candidate product based on the filtered comments.

[0238] In an optional implementation, the search submodule includes:

[0239] A first identification unit is used to identify, for each candidate product, a review tag used to describe a review topic of the high-quality review in the high-quality reviews of the candidate product;

[0240] The second determining unit is configured to determine the candidate product as a target product whose high-quality reviews are related to the search tag if the review tag includes the search tag.

[0241] In an optional implementation, the search submodule further includes:

[0242] The storage unit is used to form a corresponding table item with the comment label of the candidate product and the candidate product, and store it in the corresponding relationship between the comment label of the product review and the product.

[0243] In an optional implementation, the search submodule includes:

[0244] A first search unit, configured to search for a comment tag related to the search tag in the correspondence between the comment tags of the comments of the product and the product;

[0245] The second searching unit is used to search for a candidate commodity corresponding to the found review tag in the corresponding relationship and use it as the target commodity.

[0246] In an optional implementation, the identification submodule includes:

[0247] The second recognition unit is used to recognize, in the search text, a search tag for describing a search topic of the search text based on a search tag recognition model.

[0248] In an optional implementation, the identified submodule further includes:

[0249] A first acquisition unit, configured to acquire a sample search text set, wherein the sample search text set includes at least one sample search text annotated with a sample search tag;

[0250] The first training unit is used to train the model using the sample retrieval texts in the sample retrieval text set until all weights in the model converge to obtain the retrieval tag recognition model.

[0251] In an optional implementation, the display module includes:

[0252] A statistics submodule, used for counting the number of comments related to the search tag in the comments of the target product;

[0253] A first generating submodule, used for generating the recall reason according to the search tag and the number of comments;

[0254] The first display submodule is used to display the reason for the recall.

[0255] In an optional implementation, the generating submodule includes:

[0256] A second acquisition unit is used to acquire a recall reason generation template, wherein the recall reason generation template includes a comment quantity filling space and a search tag filling space;

[0257] The filling unit is used to fill in the number of comments in the comment number filling position in the recall reason generation template, and fill in the search tag in the search tag filling position, so as to obtain the recall reason.

[0258] In an optional implementation, the display module includes:

[0259] An acquisition submodule, used to acquire comments on the target product;

[0260] A second generating submodule, configured to generate the recall reason according to the comments of the target product;

[0261] The second display submodule is used to display the reason for the recall.

[0262] In an optional implementation, the acquisition submodule is specifically used to: count the comments related to the search tag in the comments of the target product.

[0263] In an optional implementation, the second generating submodule includes:

[0264] A generating unit is used to generate the recall reason according to the comment using the recall reason generating model.

[0265] In an optional implementation, the second generating submodule further includes:

[0266] A third acquisition unit is used to acquire multiple sample review sets, each sample review set including multiple sample reviews marked with the same sample recall reason;

[0267] The second training unit is used to train the model using sample comments in multiple sample comment sets until all weights in the model converge to obtain the recall reason generation model.

[0268] In the present application, a search text is obtained; at least candidate products whose comments are related to the search text are searched from multiple products; the candidate products are classified and screened based on the classification model to obtain the target product; the target product is recalled and the reason for recalling the target product is displayed. Among them, the comments on the product are submitted to the electronic device by the majority of users in the historical process after purchasing the product in the electronic device and using the product, that is, the comments on the product include the subjective use experience of the majority of users on the product in the historical process. Therefore, the candidate products whose comments are related to the search text are retrieved and the candidate products are classified and screened to obtain the target product, and the reason for recalling the target product is displayed. This is more convincing and credible for the current user, and more friendly to the current user, thereby increasing the possibility of the retrieved target product being purchased by the current user, and further improving the user's purchasing experience. Furthermore, displaying the reason for recall can increase the user's trust in the search method and increase the user's stickiness.

[0269] The embodiment of the present application also provides a non-volatile readable storage medium, which stores one or more modules (programs). When the one or more modules are applied to a device, the device can execute instructions (instructions) of each method step in the embodiment of the present application.

[0270] The embodiment of the present application provides one or more machine-readable media on which instructions are stored, and when executed by one or more processors, the electronic device executes the commodity retrieval method described in one or more of the above embodiments. In the embodiment of the present application, the electronic device includes a server, a gateway, a sub-device, etc., and the sub-device is an Internet of Things device or other device.

[0271] The embodiments of the present disclosure may be implemented as an apparatus configured as desired using any appropriate hardware, firmware, software, or any combination thereof, and the apparatus may include electronic devices such as servers (clusters), terminal devices such as IoT devices, and the like.

[0272] Figure 7 An exemplary apparatus 1300 that can be used to implement various embodiments described in this application is schematically shown.

[0273] For one embodiment, Figure 7 An exemplary apparatus 1300 is shown having one or more processors 1302, a control module (chip set) 1304 coupled to at least one of the (one or more) processors 1302, a memory 1306 coupled to the control module 1304, a non-volatile memory (NVM) / storage device 1308 coupled to the control module 1304, one or more input / output devices 1310 coupled to the control module 1304, and a network interface 1312 coupled to the control module 1306.

[0274] The processor 1302 may include one or more single-core or multi-core processors, and the processor 1302 may include any combination of general-purpose processors or special-purpose processors (such as graphics processors, application processors, baseband processors, etc.). In some embodiments, the device 1300 can be used as a server device such as a gateway or controller described in the embodiments of the present application.

[0275] In some embodiments, the device 1300 may include one or more computer-readable media (e.g., memory 1306 or NVM / storage device 1308) having instructions 1314 and one or more processors 1302 configured to execute the instructions 1314 in combination with the one or more computer-readable media to implement a module to perform the actions described in the present disclosure.

[0276] For one embodiment, the control module 1304 may include any suitable interface controller to provide any suitable interface to at least one of the processor(s) 1302 and / or any suitable device or component in communication with the control module 1304 .

[0277] The control module 1304 may include a memory controller module to provide an interface to the memory 1306. The memory controller module may be a hardware module, a software module, and / or a firmware module.

[0278] The memory 1306 may be used, for example, to load and store data and / or instructions 1314 for the device 1300. For one embodiment, the memory 1306 may include any suitable volatile memory, such as a suitable DRAM. In some embodiments, the memory 1306 may include double data rate type four synchronous dynamic random access memory (DDR4 SDRAM).

[0279] For one embodiment, control module 1304 may include one or more input / output controllers to provide an interface to NVM / storage device 1308 and input / output device(s) 1310 .

[0280] For example, NVM / storage 1308 may be used to store data and / or instructions 1314. NVM / storage 1308 may include any suitable non-volatile memory (e.g., flash memory) and / or may include any suitable non-volatile storage device(s) (e.g., one or more hard disk drives (HDDs), one or more compact disk (CD) drives, and / or one or more digital versatile disk (DVD) drives).

[0281] NVM / storage device 1308 may include storage resources that are physically part of the device on which apparatus 1300 is installed, or it may be accessible to the device without being part of the device. For example, NVM / storage device 1308 may be accessed via input / output device(s) 1310 over a network.

[0282] (One or more) input / output devices 1310 may provide an interface for the apparatus 1300 to communicate with any other appropriate device, and the input / output device 1310 may include a communication component, an audio component, a sensor component, etc. The network interface 1312 may provide an interface for the apparatus 1300 to communicate through one or more networks, and the apparatus 1300 may wirelessly communicate with one or more components of a wireless network according to any of one or more wireless network standards and / or protocols, for example, accessing a wireless network based on a communication standard, such as WiFi, 2G, 3G, 4G, 5G, etc., or a combination thereof for wireless communication.

[0283] For one embodiment, at least one of the processor(s) 1302 may be packaged together with the logic of one or more controllers (e.g., a memory controller module) of the control module 1304. For one embodiment, at least one of the processor(s) 1302 may be packaged together with the logic of one or more controllers of the control module 1304 to form a system-in-package (SiP). For one embodiment, at least one of the processor(s) 1302 may be integrated on the same die with the logic of one or more controllers of the control module 1304. For one embodiment, at least one of the processor(s) 1302 may be integrated on the same die with the logic of one or more controllers of the control module 1304 to form a system-on-chip (SoC).

[0284] In various embodiments, the device 1300 may be, but is not limited to, a terminal device such as a server, a desktop computing device, or a mobile computing device (e.g., a laptop computing device, a handheld computing device, a tablet computer, a netbook, etc.). In various embodiments, the device 1300 may have more or fewer components and / or a different architecture. For example, in some embodiments, the device 1300 includes one or more cameras, a keyboard, a liquid crystal display (LCD) screen (including a touch screen display), a non-volatile memory port, multiple antennas, a graphics chip, an application-specific integrated circuit (ASIC), and a speaker.

[0285] An embodiment of the present application provides an electronic device, including: one or more processors; and one or more machine-readable media having instructions stored thereon, which, when executed by the one or more processors, enable the processor to execute the commodity retrieval method as described in one or more of the embodiments of the present application.

[0286] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0287] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0288] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0289] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0290] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0291] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present application.

[0292] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0293] The above is a detailed introduction to a product retrieval method and device provided by the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for general technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A commodity search method, characterized in that: include: Get the search text; Retrieving at least one candidate product whose review is related to the search text from among the plurality of products; Classify and screen the candidate products based on the classification model to obtain target products; Recall the target product and present the reason for recalling the target product; The classifying and screening the candidate commodities based on the classification model to obtain the target commodities includes: Determine the quality reviews for each candidate product based on the classification model; At least one target product whose high-quality reviews are related to the search tag is searched among the candidate products.

2. The method according to claim 1, characterized in that The step of searching at least one candidate product whose comments are related to the search text from among the multiple products includes: In the search text, identifying a search tag for describing a search subject of the search text; At least a candidate product whose review is related to the search tag is searched for among a plurality of products.

3. The method according to claim 1, characterized in that The method of determining the high-quality reviews of each candidate product based on the classification model includes: For each candidate product, comments from high-quality users are selected from the comments on the candidate product based on the classification model, and high-quality comments on the candidate product are determined based on the comments from the high-quality users based on the classification model.

4. The method according to claim 3, characterized in that The selecting, based on the classification model, the comments of high-quality users from the comments of the candidate products comprises: Determining high-quality users among users who comment on the candidate product based on the classification model; Obtain the comments of the high-quality users on the candidate products.

5. The method according to claim 4, characterized in that The step of determining high-quality users from users who comment on the candidate products based on the classification model includes: For each user who comments on the candidate product, the user's historical operational reputation, account points and historical interactivity are obtained; based on the classification model, the user's quality score is obtained according to the historical operational reputation, account points and historical interactivity; if the user's quality score is greater than a first preset threshold, the user is determined to be a high-quality user.

6. The method according to claim 3, characterized in that The method of determining the high-quality reviews of the candidate product based on the reviews of high-quality users based on the classification model includes: Select reviews from high-quality users that include at least one of product information, logistics information, and merchant information; Determine the high-quality reviews of the candidate product based on the selected reviews based on the classification model.

7. The method according to claim 6, characterized in that The method of determining the high-quality reviews of the candidate product based on the selected reviews based on the classification model includes: Obtain the semantic fluency of each selected comment based on the classification model; Among the selected comments, screening comments whose semantic fluency is greater than or equal to a second preset threshold; Determine high-quality reviews for the candidate product based on the filtered reviews.

8. The method according to claim 7, characterized in that The obtaining of the semantic fluency of each selected comment based on the classification model includes: Use the language model LM to obtain the semantic fluency of each comment.

9. The method according to claim 8, characterized in that The method further comprises: A sample comment set is obtained, wherein the sample comment set includes at least one sample comment annotated with semantic coherence; and a model is trained using the sample comments in the sample comment set until all weights in the model converge to obtain the language model.

10. The method according to claim 6, characterized in that The method of determining the high-quality reviews of the candidate product based on the selected reviews based on the classification model includes: Obtaining the relevance between each selected review and the candidate product based on the classification model; Among the selected reviews, screening reviews whose relevance to the candidate product is greater than or equal to a third preset threshold; Determine high-quality reviews for the candidate product based on the filtered reviews.

11. The method according to claim 10, characterized in that The step of obtaining the relevance between each selected review and the candidate product based on the classification model includes: Obtain a topic for describing the industry to which the candidate product belongs; For each review, the relevance between the review and the theme of the industry is obtained based on the text topic model, and is used as the relevance between the review and the candidate product.

12. The method according to claim 6, characterized in that The method of determining the high-quality reviews of the candidate product based on the selected reviews based on the classification model includes: Obtaining the semantic fluency of each selected comment based on the classification model, and obtaining the relevance between each selected comment and the candidate product based on the classification model; Among the selected reviews, screening reviews whose semantic coherence is greater than or equal to a second preset threshold and whose relevance to the candidate product is greater than or equal to a third preset threshold; Determine high-quality reviews for the candidate product based on the filtered reviews.

13. The method according to claim 1, characterized in that The step of searching at least one target product among candidate products, whose high-quality reviews are related to the search tag, includes: For each candidate product, identifying a review tag for describing a review topic of the high-quality review in the high-quality reviews of the candidate product; If the review tag includes the search tag, the candidate product is determined as a target product whose high-quality reviews are related to the search tag.

14. The method according to claim 13, characterized in that The method further comprises: The comment labels of the candidate commodities and the candidate commodities are formed into corresponding table items, and stored in the corresponding relationship between the comment labels of the commodity reviews and the commodities.

15. The method according to claim 14, characterized in that The step of searching at least one target product among candidate products, whose high-quality reviews are related to the search tag, includes: In the correspondence between the comment tags of the product reviews and the product, searching for the comment tags related to the search tag; A candidate product corresponding to the found review tag is searched in the corresponding relationship and used as the target product.

16. The method according to claim 2, characterized in that The step of identifying, in the search text, a search tag for describing a search subject of the search text comprises: In the search text, a search tag for describing a search subject of the search text is identified based on a search tag identification model.

17. The method according to claim 16, characterized in that The method further comprises: Acquire a sample search text set, wherein the sample search text set includes at least one sample search text annotated with a sample search tag; The model is trained using the sample retrieval texts in the sample retrieval text set until all weights in the model converge, thereby obtaining the retrieval tag recognition model.

18. The method according to claim 2, characterized in that The reason for recalling the target product includes: Among the reviews of the target product, counting the number of reviews related to the search tag; Generate the recall reason according to the search tag and the number of comments; Present the reason for the recall.

19. The method according to claim 18, characterized in that The generating the recall reason according to the search tag and the number of comments includes: Obtaining a recall reason generation template, wherein the recall reason generation template includes a comment quantity filling space and a search tag filling space; The number of comments is filled in the comment number filling space in the recall reason generation template, and the search tag is filled in the search tag filling space, so as to obtain the recall reason.

20. The method according to claim 2, characterized in that The reason for recalling the target product includes: Obtaining reviews of the target product; generating the recall reason according to the review of the target product; Present the reason for the recall.

21. The method according to claim 20, characterized in that The step of obtaining the comments of the target product includes: Among the reviews of the target product, the reviews related to the search tag are counted.

22. The method according to claim 20, characterized in that Generating the recall reason according to the comment includes: The recall reason generation model is used to generate the recall reason according to the review.

23. The method according to claim 22, characterized in that The method further comprises: Obtain multiple sample review sets, each sample review set including multiple sample reviews annotated with the same sample recall reason; The model is trained using sample comments in multiple sample comment sets until all weights in the model converge, thereby obtaining the recall reason generation model.

24. A commodity search device, characterized in that: include: The acquisition module is used to obtain the search text; A search module, used to search at least one candidate product whose comments are related to the search text from among a plurality of products; A screening module, used to classify and screen the candidate commodities based on a classification model to obtain target commodities; A recall module, used for recalling the target product; A display module, used to display the reason for recall of the target product; The screening module comprises: A determination submodule is used to determine the quality reviews of each candidate product based on the classification model; The search submodule is used to search at least one target product whose high-quality reviews are related to the search tag among the candidate products.

Images