Commodity Search Category Identification Method and Its Device, Equipment, Medium, and Product

Abstract

This application relates to a method and apparatus, device, medium, and product for identifying product search categories in the field of e-commerce information technology. The method includes: receiving a product search request submitted by a user to an online store and obtaining the query term carried by the request; constructing joint coding information, where the joint coding information includes the literal feature information of the query term, the personal feature information of the user, the store feature information of the online store, and the augmented feature information of the query term; constructing the combined feature information by combining the literal feature information with the fusion feature information obtained through deep feature interaction of the joint coding information; and performing classification based on the deep semantic information of the combined feature information to obtain the category in the product classification system to which the query term classification is mapped. In this application, the reference information obtained during the process of predicting the query term classification is more abundant, and a more accurate prediction result can be obtained.

Description

Technical Field

[0001] This application relates to the field of e-commerce information technology, and in particular to a method for identifying product search categories, as well as corresponding devices, computer equipment, computer-readable storage media, and computer program products. Background Art

[0002] In the e-commerce search scenario, by having a user input a query term to identify the user's product category intention and recall products that match the user's product category intention, it can better meet the user's query purpose.

[0003] For example, when a user inputs "Guangming" for query, among the corresponding obtained product data, some of the categories are "milk" and some are "rice". According to the search historical behavior data, among the users who search for "Guangming", the number of users who click on products in the "milk" category is much larger than the number of users who click on products in the "rice" category. In this case, the category prediction model will give such a prediction result: the relevance of the "milk" category to "Guangming" is higher than the relevance of the "rice" category to "Guangming". Therefore, products in the "milk" category will be preferentially recommended when sorting, thus improving the business value of the search.

[0004] In reality, when a user submits a query term as a keyword, it is usually a short input of a single word, generally two or three Chinese characters or two or three English words, that is, the query term is a short text, and the information it contains is very low. Therefore, it is very difficult to accurately identify the user's product category intention through a short text, and the prediction accuracy of directly using short text modeling in the industry is not high.

[0005] In view of this, in response to the need to improve the exploration results to hit the user's shopping intention in the e-commerce search scenario, the applicant has made corresponding explorations. Summary of the Invention

[0006] The purpose of this application is to solve at least one of the above problems and provide a method for identifying product search categories, as well as corresponding devices, computer equipment, computer-readable storage media, and computer program products.

[0007] To meet the various purposes of this application, the following technical solutions are adopted:

[0008] On the one hand, to meet one of the purposes of this application, a method for identifying product search categories is provided, including the following steps:

[0009] Receive a product search request submitted by a user to an online store and obtain the query term carried by the request;

[0010] Construct joint encoded information, where the joint encoded information includes the literal feature information of the query term, the personal feature information of the user, the store feature information of the online store, and the augmented feature information of the query term;

[0011] Construct the literal feature information combined with the fused feature information obtained by deep feature interaction of the joint encoded information into comprehensive feature information;

[0012] Classify according to the deep semantic information of the comprehensive feature information to obtain the category in the commodity classification system mapped by the query term classification.

[0013] In some specific embodiments, constructing joint encoded information includes the following steps:

[0014] Perform word embedding on the query term to vectorize and encode it into literal feature information;

[0015] Match the augmented vocabulary set composed of the related words of the query term from a preset word list, and vectorize and encode the augmented vocabulary set into augmented feature information, where the related words include the synonyms and / or co-occurrence words of the query term;

[0016] Call multiple feature data of the user and vectorize and encode the feature data into personal feature information;

[0017] Call multiple feature data of the online store and vectorize and encode the feature data into store feature information;

[0018] Perform multi-channel splicing on the literal feature information, augmented feature information, personal feature information, and store feature information to construct joint encoded information.

[0019] In some specific embodiments, matching the augmented vocabulary set composed of the related words of the query term from a preset word list and vectorizing and encoding the augmented vocabulary set into augmented feature information includes the following steps:

[0020] Semantically match the synonyms of the query term from a preset synonym list as its related words;

[0021] Semantically match the co-occurrence words of the query term from a preset co-occurrence word list as its related words;

[0022] Construct the synonyms and co-occurrence words into the augmented vocabulary set of the query term;

[0023] According to a preset encoding word list, convert the related words in the augmented vocabulary set into codes for vectorization to obtain the augmented feature information of the query term.

[0024] In some of the enhanced embodiments, constructing the fused feature information obtained by performing deep feature interaction on the literal feature information in combination with the joint coding information into comprehensive feature information includes the following steps:

[0025] Performing deep feature interaction on the joint coding information using an attention layer to obtain fused feature information;

[0026] Using a splicing layer to perform multi-channel splicing on the fused feature information and the literal feature information of the query term to obtain comprehensive feature information.

[0027] In some of the enhanced embodiments, classifying according to the deep semantic information of the comprehensive feature information to obtain the category in the commodity classification system to which the query term classification is mapped includes the following steps:

[0028] Using a text feature extraction model to extract deep semantic information from the comprehensive feature information;

[0029] Using a classifier to map the deep semantic information to the corresponding classification space of a preset commodity classification system to obtain the confidence corresponding to each classification;

[0030] Determining the corresponding category in the commodity classification system for the query term according to the classification with the maximum confidence.

[0031] In some of the extended embodiments, after the step of classifying according to the deep semantic information of the comprehensive feature information to obtain the category in the commodity classification system to which the query term classification is mapped, the following steps are included:

[0032] Retrieving from the commodity database of the online store the commodity data whose commodity information matches the query term to form a commodity candidate list;

[0033] Sorting the commodity candidate list by category according to the confidence obtained for each classification in the classifier classification space to obtain a commodity recommendation list;

[0034] Pushing the commodity recommendation list to the user to respond to the commodity search request.

[0035] In some of the extended embodiments, after the step of classifying according to the deep semantic information of the comprehensive feature information to obtain the category in the commodity classification system to which the query term classification is mapped, the following steps are included:

[0036] Retrieving from the commodity database of the online store the commodity data whose commodity information matches the query term and belongs to the category with the maximum confidence to form a commodity recommendation list;

[0037] Pushing the commodity recommendation list to the user to respond to the commodity search request.

[0038] On the other hand, a product search category recognition device provided to meet one of the purposes of this application includes a request receiving module, a coding construction module, a feature integration module, and a category recognition module, where: the request receiving module is used to receive a product search request submitted by a user to an online store and obtain the query term carried by the request; the coding construction module is used to construct combined coding information, which includes the literal feature information of the query term, the personal feature information of the user, the store feature information of the online store, and the augmented feature information of the query term; the feature integration module is used to construct the literal feature information and the fused feature information obtained through deep feature interaction of the combined coding information into comprehensive feature information; the category recognition module is used to classify according to the deep semantic information of the comprehensive feature information and obtain the category in which the query term classification is mapped to the product classification system.

[0039] In some deepened embodiments, the coding construction module includes: a literal coding unit, which is used to perform word embedding on the query term to vectorize and code it into literal feature information; an augmented coding unit, which is used to match the augmented vocabulary set composed of the related words of the query term from a preset word list and vectorize and code the augmented vocabulary set into augmented feature information, where the related words include the synonyms and / or co-occurring words of the query term; a user coding unit, which is used to call multiple feature data of the user and vectorize and code the feature data into personal feature information; a store coding unit, which is used to call multiple feature data of the online store and vectorize and code the feature data into store feature information; a combined processing unit, which is used to perform multi-channel splicing on the literal feature information, augmented feature information, personal feature information, and store feature information to construct combined coding information.

[0040] In some specific embodiments, the augmented coding unit includes: a synonym matching subunit, which is used to semantically match the synonyms of the query term from a preset synonym list as its related words; a co-occurrence matching subunit, which is used to semantically match the co-occurring words of the query term from a preset co-occurrence word list as its related words; a set construction subunit, which is used to construct the synonyms and co-occurring words into the augmented vocabulary set of the query term; a vector conversion subunit, which is used to convert the related words in the augmented vocabulary set into codes according to a preset coding word list to achieve vectorization and obtain the augmented feature information of the query term.

[0041] In some deepened embodiments, the feature integration module includes: a feature interaction unit configured to perform deep feature interaction on the joint encoded information by using an attention layer to obtain fused feature information; and a splicing and integration unit configured to perform multi-channel splicing on the fused feature information and the literal feature information of the query term by using a splicing layer to obtain integrated feature information.

[0042] In some deepened embodiments, the category recognition module includes: a feature extraction unit configured to extract deep semantic information from the integrated feature information by using a text feature extraction model; a classification mapping unit configured to map the deep semantic information to a classification space corresponding to a preset product classification system by using a classifier to obtain the confidence levels corresponding to each classification therein; and a category determination unit configured to determine the corresponding category in the product classification system for the query term according to the classification with the highest confidence level.

[0043] In some extended embodiments, subsequent to the category recognition module, it includes: a full-category retrieval module configured to retrieve product data whose product information matches the query term from the product database of the online store to form a product candidate list; a category ranking module configured to rank the product candidate list by category according to the confidence levels obtained for each classification in the classifier's classification space to obtain a product recommendation list; and a list push module configured to push the product recommendation list to the user to respond to the product search request.

[0044] In some extended embodiments, subsequent to the category recognition module, it includes: a category retrieval module configured to retrieve product data whose product information matches the query term and belongs to the category with the highest confidence level from the product database of the online store to form a product recommendation list; and a list push module configured to push the product recommendation list to the user to respond to the product search request.

[0045] On the other hand, a computer device provided to meet one of the purposes of the present application includes a central processing unit and a memory, and the central processing unit is configured to call and run a computer program stored in the memory to execute the steps of the product search category recognition method of the present application.

[0046] On the other hand, a computer-readable storage medium provided to meet another purpose of the present application stores a computer program implemented based on the product search category recognition method in the form of computer-readable instructions. When the computer program is called and run by a computer, it executes the steps included in the method.

[0047] On the other hand, a computer program product provided to meet another purpose of the present application includes a computer program / instructions, and when the computer program / instructions are executed by a processor, the steps of the method in any one of the embodiments of the present application are implemented.

[0048] Compared with the prior art, the present application has multiple advantages, including at least the following aspects:

[0049] Based on the literal feature information of the query term when the user submits a product search request in the e-commerce search scenario, the present application comprehensively constructs joint coding information by integrating the user's personal feature information, the store feature information of the online store, and the augmented feature information expanded according to the query term. The joint coding information contains both the personal information features of the user and the augmented information features composed of the associated words of the online store and the query term. After performing deep feature interaction on the joint coding information, the weight values of the query term and other information features are extracted, which can measure the features related to the query term. Finally, these related features are combined with the literal feature information of the query term again to obtain comprehensive feature information. Based on the comprehensive feature information, classification mapping of the corresponding product classification system is performed to determine the product category that matches the query term, thereby realizing the prediction of the user's product query intention. Since the associated words of the query term and the features of the user and the online store are fully utilized in the process of determining the product category, more abundant reference information is available during the classification prediction process, and a more accurate prediction result can be obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The above and / or additional aspects and advantages of the present application will become apparent and easy to understand from the following description of the embodiments in conjunction with the drawings, where:

[0051] Figure 1 is a schematic flowchart of a typical embodiment of the product search category recognition method of the present application;

[0052] Figure 2 is a schematic flowchart of the process of constructing joint coding information in the embodiment of the present application;

[0053] Figure 3 is a schematic flowchart of the process of constructing the augmented feature information of the query term in the embodiment of the present application;

[0054] Figure 4 is a schematic flowchart of the process of comprehensively obtaining comprehensive feature information by combining the fused feature information with the literal feature information of the query term in the embodiment of the present application;

[0055] Figure 5 is a schematic diagram of the network structure for implementing the encoding process of the comprehensive feature information according to the present application;

[0056] Figure 6 is a schematic flowchart of the process of predicting the product category corresponding to the query term according to the comprehensive feature information in the embodiment of the present application;

[0057] Figure 7 It is a schematic diagram of the network structure of the exemplary hierarchical multi-label classification network of the present application;

[0058] Figure 8 It is a schematic flowchart of the process of sorting the search results of query terms according to the confidence of each classification of the classifier to obtain a commodity recommendation list in the embodiment of the present application;

[0059] Figure 9 It is a schematic flowchart of the process of obtaining a commodity recommendation list according to the category corresponding to the maximum confidence of the classifier to constrain the query term search results in the embodiment of the present application;

[0060] Figure 10 It is a principle block diagram of the commodity search category recognition device of the present application;

[0061] Figure 11 It is a schematic diagram of the structure of a computer device adopted by the present application. Detailed implementation manners

[0062] The embodiments of the present application will be described in detail below. The examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions from beginning to end. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be construed as a limitation of the present application.

[0063] Those skilled in the art of the present technology can understand that, unless specifically stated otherwise, the singular forms "a", "an", "the" and "said" used herein may also include the plural forms. It should be further understood that the term "including" used in the specification of the present application means the presence of the described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or their groups. It should be understood that when we say that an element is "connected" or "coupled" to another element, it can be directly connected or coupled to other elements, or there may also be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The phrase "and / or" used herein includes all or any unit and all combinations of one or more related listed items.

[0064] Those skilled in the art of the present technology can understand that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as the general understanding of those of ordinary skill in the art to which the present application belongs. It should also be understood that terms such as those defined in a general dictionary should be understood to have a meaning consistent with the meaning in the context of the prior art, and will not be interpreted with an idealized or overly formal meaning unless specifically defined as here.

[0065] Those skilled in the art of the present technology can understand that the "client", "terminal", and "terminal device" used herein include both devices with a wireless signal receiver that only has the ability to receive without transmitting, and devices with receiving and transmitting hardware that have the receiving and transmitting hardware capable of two-way communication on a two-way communication link. Such devices can include: cellular or other communication devices such as personal computers, tablet computers, etc., which have a single-line display or a multi-line display or a cellular or other communication device without a multi-line display; PCS (Personal Communications Service), which can combine voice, data processing, fax, and / or data communication capabilities; PDA (Personal Digital Assistant), which can include a radio frequency receiver, a pager, Internet / intranet access, a web browser, a notepad, a calendar, and / or a GPS (Global Positioning System) receiver; conventional laptop and / or palm-held computers or other devices, which are conventional laptop and / or palm-held computers or other devices with and / or including a radio frequency receiver. The "client", "terminal", and "terminal device" used herein can be portable, transportable, installed in a vehicle (air, sea, and / or land), or suitable for and / or configured to run locally, and / or run in a distributed form at any other location on the earth and / or in space. The "client", "terminal", and "terminal device" used herein can also be a communication terminal, an Internet access terminal, a music / video playback terminal, for example, it can be a PDA, a MID (Mobile Internet Device), and / or a mobile phone with music / video playback function, or can also be a smart TV, a set-top box, etc.

[0066] The hardware referred to by names such as "server", "client", and "service node" in this application is essentially an electronic device with the equivalent capabilities of a personal computer, and is a hardware device with the necessary components disclosed by the von Neumann principle, including a central processing unit (including an arithmetic unit and a controller), a memory, an input device, and an output device. The computer program is stored in its memory, and the central processing unit loads the program stored in the external memory into the memory for execution, executes the instructions in the program, and interacts with the input / output devices to complete specific functions.

[0067] It should be noted that the concept of "server" referred to in this application can similarly be extended to the case of server clusters. According to the network deployment principles understood by those skilled in the art, the various servers should be logically divided. Physically, these servers can either be independent of each other but can be invoked through interfaces, or integrated into a single physical computer or a set of computer clusters. Those skilled in the art should understand this flexibility and should not be restricted by this when implementing the network deployment method of this application.

[0068] One or several technical features of this application, unless explicitly specified, can either be deployed on the server and accessed by the client remotely invoking the online service interface provided by the server, or directly deployed and run on the client for access.

[0069] The neural network models cited or possibly cited in this application, unless explicitly specified, can either be deployed on a remote server and remotely invoked by the client, or deployed on a client capable of handling the device capabilities for direct invocation. In some embodiments, when it runs on the client, its corresponding intelligence can be obtained through transfer learning to reduce the requirements for the client's hardware operating resources and avoid excessive occupation of the client's hardware operating resources.

[0070] All kinds of data involved in this application, unless explicitly specified, can either be remotely stored on the server or stored on the local terminal device, as long as it is suitable for being invoked by the technical solution of this application.

[0071] Those skilled in the art should be aware that although the various methods of this application are described based on the same concept and thus show commonality with each other, unless otherwise specified, these methods can all be executed independently. Similarly, for the various embodiments disclosed in this application, they are all proposed based on the same inventive concept. Therefore, for concepts with the same expression, as well as concepts that are only appropriately transformed for convenience although the concept expressions are different, they should be equivalently understood.

[0072] For the various embodiments to be disclosed in this application, unless explicitly pointed out that there is a mutually exclusive relationship between them, the relevant technical features involved in each embodiment can be cross-combined to flexibly construct new embodiments, as long as this combination does not deviate from the creative spirit of this application and can meet the requirements in the prior art or solve certain deficiencies in the prior art. Those skilled in the art should be aware of this flexibility.

[0073] A method for identifying product search categories in this application can be programmed as a computer program product and deployed to run on a client or a server. For example, in an exemplary application scenario of this application, it can be deployed and implemented on the server of an e-commerce platform. Thus, by accessing the interface opened after the computer program product runs, human-computer interaction can be performed with the process of the computer program product through a graphical user interface to execute this method.

[0074] Please refer to Figure 1 , in a typical embodiment of the method for identifying product search categories in this application, the following steps are included:

[0075] Step S1100: Receive a product search request submitted by a user to an online store, and obtain the query term carried by this request:

[0076] In an exemplary e-commerce scenario, the e-commerce platform provides an independent website service. Merchants rent the independent website of the e-commerce platform, deploy an online store, and list product information on the online store for the user, that is, the consumer user, to select and purchase. To facilitate the consumer user to query the products in the online store, the e-commerce platform also configures a product search service for the online store. The consumer user can input a search keyword through the product search function provided by the online store. The keyword is the query term, and the query term is submitted to the product search service to query and obtain the product information of the products that match the query term in the online store.

[0077] After the consumer user inputs the keyword and confirms the submission, the page program packages the keyword as a query term and submits it as a product search request to the product search service. After receiving the product search request, the independent site server that provides the product search service parses it to obtain the query term therein.

[0078] In a modified embodiment based on this, text preprocessing can be performed on the keyword in the page program of the consumer user's terminal device to form the query term, or the query term parsed and obtained from the product search request can be subjected to the text preprocessing on the server to clean the query term. The text preprocessing can include conventional operations such as removing stop words and removing spaces that do not change the meaning expression of the vocabulary.

[0079] Step S1200: Construct combined coding information, which includes the literal feature information of the query term, the personal feature information of the user, the store feature information of the online store, and the augmented feature information of the query term:

[0080] To obtain the associated features corresponding to the query term, various basic information that has a potential association with the query term can be integrated, and the feature information corresponding to the literal meaning of the query term can be constructed together as joint coding information. Then, after performing deep semantic processing on the joint coding information, deeper semantic feature information can be obtained, so as to provide the associated features for the query term through the obtained deeper semantic feature information, thereby being used to expand the information representation range corresponding to the query term and serving as richer basic information required for subsequent classification.

[0081] The basic information that has a potential association with the query term mainly includes, in this embodiment, personal feature information corresponding to the personal features of the consumer user, store feature information corresponding to the store features of the online store, and augmented feature information related to the semantics of the query term.

[0082] The personal feature information includes, but is not limited to, any combination of the following information of the user: user ID, user age, user gender, user preference tags, etc. Generally speaking, any information suitable for describing the personal features of the consumer user can be used to construct their personal feature information.

[0083] The store feature information includes, but is not limited to, any combination of the following information of the online store: store ID, main product category tags, store name, service area of the store, etc. Generally speaking, any information suitable for describing the store features of the online store can be used to construct their store feature information.

[0084] The augmented feature information is the information obtained by expanding according to the meaning of the query term, and can include associated word information such as synonyms and / or co-occurrence words corresponding to the query term. These synonyms and co-occurrence words are pre-statistically generated and can be retrieved through fuzzy rule matching or semantic matching methods.

[0085] In one embodiment, when obtaining the associated words of the query term, it can first be detected whether the query term is longer than a preset number of characters, such as two single characters (words) or three single characters (words). When it is longer than the preset number of characters, it is not excluded that there are multiple words in the query term. In this case, the query term can be segmented to obtain its corresponding segmentation set, and then, according to each word in it, its synonyms and / or co-occurrence words can be obtained to make the constituent elements of the augmented feature information more abundant.

[0086] In addition to the feature information corresponding to the above various associated basic information, the query term itself can be regarded as an independent feature to form its literal meaning feature information. Among them, when the query term contains multiple segmented words, its full amount of segmented words can be used to form its literal meaning feature information. These feature information can all be represented in a vectorized form.

[0087] At this point, the original meaning feature information of the query term, the personal feature information of the user, the store feature information of the online store, and the augmented feature information of the query term can be jointly encoded to construct corresponding joint encoding information.

[0088] It is not difficult to understand that the joint encoding information is a comprehensive integration of the original meaning features of the query term, user personal features, store features, and semantic augmentation features of the query term, containing more abundant basic information, and can be used to find the associated features of the query term.

[0089] Step S1300: Construct the combined feature information by combining the original meaning feature information with the fusion feature information obtained through deep feature interaction of the joint encoding information:

[0090] For the joint encoding information mentioned above, a preset neural network model can be used to perform deep feature interaction on it, such as performing the QKV operation of the reference context, extracting the weights of other potentially associated basic information according to the original meaning feature information therein, so as to measure which feature in the joint encoding information is more relevant to the query term. Thus, the fusion feature information is obtained. That is to say, the fusion feature information gives the descriptive information corresponding to the features related to the original meaning word, that is, provides the associated features of the query term, and provides a more abundant and concentrated information reference for the intention recognition of the query term.

[0091] Accordingly, the original meaning feature information of the query term is combined with the fusion reference information to construct the combined feature information. This combined feature information realizes the integration between the query term and its associated features. It can be seen that the combined feature information obtained after semantic augmentation of the query term and comprehensive consideration of various aspects of information such as users and stores already has more abundant and comprehensive decision-making information for identifying the query intention of the user.

[0092] Step S1400: Classify according to the deep semantic information of the combined feature information to obtain the category in the commodity classification system to which the query term classification is mapped:

[0093] The comprehensive feature information includes the literal feature information of the query term and the fusion feature information, enriching the semantic information of the query term. Accordingly, a neural network model pre-trained to a convergent state can be used for representation learning to obtain its deep semantic information, and then the deep semantic information is mapped to a classification space through a fully connected layer in the subsequent classifier. The classification space is set with multiple classifications corresponding to each category of the product classification system of the online store. Accordingly, through classification mapping, the confidence levels corresponding to each classification in the classification space, that is, classification probabilities, are obtained. Among them, the category corresponding to the classification with the largest classification probability is the product category matching the query term. Thus, the prediction of the product category expected to be searched by the user according to the query term is realized. Subsequently, based on this, products can be retrieved in the corresponding categories, or the product search results corresponding to the query term can be sorted according to the confidence levels of each category, and so on.

[0094] Generally speaking, an ordinary multi-classification network can be used to classify the comprehensive feature information. To adapt to the characteristic of the product classification system of the e-commerce platform having multi-level classifications, in an adaptively modified embodiment, a hierarchical multi-label classification network (HMCN, Hierarchical Multi-Label Classification Networks) suitable for multi-level classifications can be used to implement the classification of the comprehensive feature information. Thus, the situation where the product classification system includes multiple levels can be supported. The backbone model of this hierarchical multi-label classification network can adopt any existing neural network model feasible for extracting text feature information, such as TextCNN, AlBert, etc. In addition, the hierarchical multi-label classification network can be pre-trained by those skilled in the art to a convergent state with a sufficient number of training samples according to the principles disclosed in this application and then put into use.

[0095] During training, each training sample can be collected from the search historical data generated by e-commerce platform users' searches in each online store. The keyword therein is extracted as the query term to construct the literal feature information, the store feature information, the user's personal feature information, and the augmented feature information corresponding to the query term determined according to a preset word list are constructed as the training sample. Then, according to the encoding process of this application, the comprehensive feature information of each training sample is obtained and input into the hierarchical multi-label classification network for training. In the search historical data, the category of the product that the user preferentially accesses in the result list corresponding to the query term can be used as the supervision label of this training sample to supervise the hierarchical multi-label classification network to predict the category representing the user's intention for this training sample, perform gradient update on the model, and prompt the model to converge quickly.

[0096] According to the embodiments disclosed herein, compared with the prior art, the present application has multiple advantages, including at least the following aspects:

[0097] Based on the e-commerce search scenario, the present application constructs joint coding information by comprehensively considering the literal feature information of the query term submitted by the user for a product search request, the user's personal feature information, the store feature information of the online store, and the augmented feature information extended according to the query term. The joint coding information contains both the user's personal information features and the extended information features composed of the related words of the online store and the query term. After performing deep feature interaction on the joint coding information, the weight values of the query term and other information features are extracted to measure the features related to the query term. Finally, these related features are combined with the literal feature information of the query term again to obtain comprehensive feature information. Based on the comprehensive feature information, classification mapping of the corresponding product classification system is performed to determine the product category that matches the query term, thereby realizing the prediction of the user's product query intention. Since the related words of the query term and the features of the user and the online store are fully utilized in the process of determining the product category, more reference information is available during the classification prediction process, and a more accurate prediction result can be obtained.

[0098] Please refer to Figure 2 , in some deepened embodiments, the step S1200 of constructing joint coding information includes the following steps:

[0099] Step S1210: Perform word embedding on the query term to vectorize and encode it into literal feature information:

[0100] To facilitate the processing of the query term by the neural network model, the query term can be converted into a corresponding numerical code with reference to a preset coding word table to achieve word embedding and obtain its corresponding literal feature vector to represent its literal feature information. When there are multiple segmented words in the query term, each segmented word can be correspondingly vectorized to avoid omission.

[0101] Step S1220: Match the augmented vocabulary set composed of the related words of the query term from the preset word table, and vectorize and encode the augmented vocabulary set into augmented feature information. The related words include the synonyms and / or co-occurrence words of the query term:

[0102] To expand the semantics of the query term, the synonyms and / or co-occurrence words of the query term are matched from the preset word table for providing related words as its related words, and a corresponding augmented vocabulary set is constructed. On this basis, with reference to the aforementioned coding word table, each related word in the augmented vocabulary set is word-embedded to achieve vectorization, thereby obtaining a semantically augmented feature vector as the augmented feature information corresponding to the query term.

[0103] Step S1230: Invoke multiple feature data of the user, and vectorize and encode the feature data into personal feature information:

[0104] To provide augmented information required for predicting the category intention of the query term from the user side, collect any multiple feature data from the following items of information of the user, including but not limited to: user ID, user age, user gender, user preference tags, etc. Then, vectorize these feature data correspondingly and encode them into the personal feature vector corresponding to the user to represent his personal feature information. The provision of personal feature information enriches the reference information source of the query term and facilitates providing reference features on the user side for the prediction of the query term.

[0105] Step S1240: Invoke multiple feature data of the online store, and vectorize and encode the feature data into store feature information:

[0106] To provide augmented information required for predicting the category intention of the query term from the online store side, collect any multiple feature data from the following items of information of the online store, including but not limited to: store ID, main product category tags, store name, service area of the store, etc. Then, vectorize these feature data correspondingly and encode them into the store feature vector corresponding to the user to represent its store feature information. Similarly, the provision of store feature information enriches the reference information source of the query term and facilitates providing reference features on the online store side for the prediction of the query term.

[0107] Step S1250: Perform multi-channel splicing on the literal feature information, augmented feature information, personal feature information, and store feature information to construct joint encoded information:

[0108] All the above-mentioned feature information constructed for the query term can be normalized to the same dimension. On this basis, use a splicing layer to perform multi-channel splicing on the literal feature information, augmented feature information, personal feature information, and store feature information. Thus, various feature information can be constructed into joint encoded information.

[0109] The embodiments herein disclose the encoding process of the joint encoded information of the present application. It can be seen that in addition to the literal feature information of the query term itself, semantic information is enriched for it from different channels during the encoding process. It not only considers the feature data on the user side but also the feature data on the store side. In addition, the synonymous data of the query term is also considered. Therefore, not only personalized features are added to the query term, but also augmented information features are added, forming an associated information set corresponding to the query term. Based on this, more useful reference information for predicting the category intention of the query term can be obtained through data mining.

[0110] Please refer to Figure 3, in some specific embodiments, the step S1220 of matching the associated words of the query word from a preset word list and vectorizing and encoding the augmented vocabulary set composed of the associated words into augmented feature information includes the following steps:

[0111] Step S1221, semantically matching the synonyms of the query word from a preset synonym list as its associated words:

[0112] To expand the synonyms of the query word, a synonym list can be prepared. This synonym list is usually a word list obtained by frequency statistics and optimization based on a large amount of text information on the e-commerce platform. On this basis, various methods can be used to fuzzily match one or more corresponding synonyms of the query word in the synonym list as its associated words.

[0113] In an optional embodiment, in a way of fuzzy rule matching, wildcards are applied to the query word to construct multiple query expressions, and the query expressions are used to retrieve in the synonym list, and multiple matching synonyms can be determined.

[0114] In another optional embodiment, in a way of fuzzy semantic matching, the vector representation of the query word is used to calculate the data distance with the vector representations of each synonym in the synonym list, and then the synonyms whose data distance meets the preset distance threshold are determined as the associated words of the query word.

[0115] It can be seen that no matter which method is adopted, synonyms semantically similar to the query word can be matched from the synonym list to expand the associated information of the query word.

[0116] Step S1222, semantically matching the co-occurrence words of the query word from a preset co-occurrence word list as its associated words:

[0117] To expand the co-occurrence words of the query word, a co-occurrence word list can be prepared. This co-occurrence word list is usually a word list obtained by frequency statistics and optimization of the words that co-occur with the query word based on a large amount of text information on the e-commerce platform. On this basis, various methods can be used to fuzzily match one or more corresponding co-occurrence words of the query word in the co-occurrence word list as its associated words.

[0118] In an optional embodiment, in a way of fuzzy rule matching, wildcards are applied to the query word to construct multiple query expressions, and the query expressions are used to retrieve in the co-occurrence word list, and multiple matching co-occurrence words can be determined.

[0119] In another alternative embodiment, in a manner of fuzzy semantic matching, the data distance between the vector representation of the query term and the vector representations of each co-occurring term in the co-occurrence word list is calculated, and then the co-occurring terms whose data distance meets a preset distance threshold are determined as the related terms of the query term.

[0120] It can be seen that no matter which method is adopted, co-occurring terms semantically similar to the query term can be matched from the co-occurrence word list to expand the related information of the query term.

[0121] Step S1223: Construct the synonym and co-occurring terms into an augmented vocabulary set of the query term:

[0122] The matched synonyms and co-occurring terms enrich the semantics of the query term from different perspectives. Therefore, they can be regarded as the related terms of the query term together and constructed into an augmented vocabulary set.

[0123] Step S1224: According to a preset coding word list, the related terms in the augmented vocabulary set are correspondingly converted into codes to achieve vectorization, and the augmented feature information of the query term is obtained:

[0124] Finally, similarly, according to the coding word list based on which the query term is word-embedded, each related term in the augmented vocabulary set is vectorized, so as to encode and obtain the augmented feature information corresponding to the query term.

[0125] The embodiments herein show that the synonym list and co-occurrence word list obtained by data mining on the e-commerce platform can be used to provide a rich and reliable information basis for semantic expansion of query terms, enabling query terms to provide semantic information not only through their literal meanings but also through their synonyms and co-occurring terms, providing a reliable and informative information source for subsequent provision of reference feature information to query terms.

[0126] Please refer to Figure 4 , in some in-depth embodiments, the step S1300: constructing the combined feature information by combining the literal feature information and the fusion feature information obtained by deep feature interaction of the combined coding information includes the following steps:

[0127] Step S1310: Perform deep feature interaction on the combined coding information by using an attention layer to obtain fusion feature information:

[0128] As described above, the comprehensive feature information implies the joint coding information and the literal feature information of the query term. The joint coding information contains vector representations of various associated information corresponding to amplifying semantic information for the query term. However, different query terms may be sensitive to different associated information, that is, when different associated information is combined with the query term, the implied information value is also correspondingly different. In order to discover the association relationship between the query term and these associated information, in this embodiment, as Figure 5 shown in the exemplary coding network of

[0129] Step S1320: Use a splicing layer to perform multi-channel splicing on the fusion feature information and the literal feature information of the query term to obtain comprehensive feature information:

[0130] After obtaining the fusion feature information, a splicing layer can be used to perform multi-channel splicing on the fusion feature information and the literal feature information of the query term to obtain the comprehensive feature information. Thus, the comprehensive feature information contains the semantic information of the query term itself and the semantic information of other associated information that is related to it to varying degrees, thereby providing effective reference information for the user's search category intention decision-making.

[0131] The embodiment here reveals the process of deep feature interaction of the joint coding information based on the attention layer, thereby quantifying the weights of various associated information of the query term to obtain fusion feature information, and then splicing it with the literal feature information of the query term to obtain comprehensive feature information, which can provide more reliable basic information for commodity category prediction and make commodity category prediction more accurate.

[0132] Please refer to Figure 6 , in some deepened embodiments, the step S1400: Classify according to the deep semantic information of the comprehensive feature information to obtain the category in which the query term classification is mapped to the commodity classification system, includes the following steps:

[0133] Step S1410: Use a text feature extraction model to extract deep semantic information from the comprehensive feature information:

[0134] Please combine with Figure 7 the exemplary hierarchical multi-label classification network, which uses a text feature extraction model 78 as the backbone network, followed by a classifier 70. The text feature extraction model 78 can extract its deep semantic information by performing representation learning on the comprehensive feature information input into it. In the classifier 70, it is mapped to the output layer 703 through the fully connected layer 701, and the confidence corresponding to each classification in the classification space is calculated by the output layer 703 as the classification result. As described above, the hierarchical multi-label classification network can be pre-trained with a sufficient amount of training samples until it converges and then put into use.

[0135] Step S1420: Use the classifier to map the deep semantic information to the classification space corresponding to the preset product classification system, and obtain the confidence corresponding to each classification therein:

[0136] The classifier fully connects the deep semantic information through the fully connected layer and then maps it to the classification space corresponding to the product classification system of the online store, thereby obtaining the confidence corresponding to each classification mapped to the classification space.

[0137] Step S1430: Determine the corresponding category in the product classification system for the query term according to the classification with the highest confidence:

[0138] In order to predict the search intent category of the query term carried in the product search request, the category corresponding to the classification with the highest confidence in the classification space can be determined as the product category corresponding to the query term, that is, the product category that the product search request expects to search for.

[0139] The embodiments herein exemplarily disclose the process by which the hierarchical multi-label classification network obtains deep semantic information based on the comprehensive feature information and makes classification predictions. It can be seen that after being encoded by the present application, the hierarchical multi-label classification network can obtain richer and more reliable basic information, that is, the comprehensive feature information, for making a preliminary judgment on the user's search intent category for the product search request, so as to hopefully improve the hit rate of the product search results hitting the user's search intent category.

[0140] Please refer to Figure 8 , in the extended partial embodiments, after the step S1400: Classify according to the deep semantic information of the comprehensive feature information to obtain the step of the category to which the query term classification is mapped in the product classification system, the following steps are included:

[0141] Step S2100: Retrieve the product data whose product information matches the query term from the product database of the online store to form a product candidate list:

[0142] As an example of product search, in this embodiment, first, according to the query term, product data whose product information in the product database of the online store matches the query term is retrieved from the product database of the online store, and these product data form a product candidate list, where each product corresponds to one piece of product data, which can be represented as a data record in this list.

[0143] The product information used for retrieval here can be the product title and / or product details in the product data. The retrieval method used can be any one of exact rule matching, fuzzy rule matching, fuzzy semantic matching, etc. Among the obtained product data, there may be product data corresponding to multiple categories in the product classification system.

[0144] Step S2200: Sort the product candidate list by category according to the confidence levels obtained for each classification in the classifier's classification space to obtain a product recommendation list:

[0145] When the neural network model of this application has predicted the confidence levels for each category in the product classification system corresponding to the query term, the confidence levels of each classification in the classification space of the corresponding classifier can be used to sort each piece of product data in the product candidate list according to the category confidence level. Thus, a product recommendation list is obtained.

[0146] Step S2300: Push the product recommendation list to the user to respond to the product search request:

[0147] Finally, the product recommendation list is pushed to the user to complete the response to the product search request. Of course, before pushing the product recommendation list to the user, the product data in the product recommendation list can also be optimized or truncated to streamline the search results. From this, it can be known that products in categories with higher confidence levels can be preferentially recommended to the user. Since this category is the result of decision-making based on the query term and its various associated information, it can be expected that the preferentially recommended product data can better meet the user's expectations and achieve a more accurate search recommendation result.

[0148] The embodiments here disclose sorting the results obtained from the search according to the query term based on the confidence levels of each category in the product classification system predicted by the neural network model, so that product data that meets the user's search intention is preferentially recommended for the user to view. On the one hand, the search accuracy is improved to enhance the user experience. On the other hand, the exposure rate of products in highly relevant valuable categories can also be increased, making the search recommendation strategy of the online store more reasonable, thereby comprehensively enhancing the economic benefits of the online store.

[0149] Please refer to Figure 9, in the extended partial embodiments, after the step S1400 of classifying according to the deep semantic information of the comprehensive feature information to obtain the category in the commodity classification system to which the query word classification is mapped, the following steps are included:

[0150] Step S3100: Retrieve the commodity data whose commodity information in the commodity database of the online store matches the query word and belongs to the category with the highest confidence, and form a commodity recommendation list:

[0151] As an example of commodity search, in this embodiment, according to the query word, retrieve the commodity data whose commodity information in the commodity database of the online store matches the query word, and constrain the category of the commodity to be searched during the search to the category corresponding to the highest confidence classified by the neural network model of the present application. Then, form a commodity recommendation list with these commodity data, where each commodity corresponds to one commodity data, which can be represented as a data record in this list.

[0152] Similarly, the commodity information used for retrieval here can be the commodity title and / or commodity details in the commodity data. The retrieval method adopted can be any one of precise rule matching, fuzzy rule matching, fuzzy semantic matching, etc. Among the obtained commodity data, only the commodity data corresponding to the category with the highest confidence in the commodity classification system is included.

[0153] Step S3200: Push the commodity recommendation list to the user to respond to the commodity search request:

[0154] Similarly, push the commodity recommendation list to the user to complete the response to the commodity search request. Before pushing the commodity recommendation list to the user, the commodity data in the commodity recommendation list can also be optimized or truncated to streamline the search results. It can be known from this that since the commodity recommendation list only contains the commodities corresponding to the categories with relatively high confidence and is relatively pure, it can be expected that the preferably recommended commodity data can better meet the user's expectations and achieve relatively accurate search recommendation results.

[0155] The embodiment here reveals that when searching for commodity data for the user, only retrieve the commodity data that matches the search of the query word for the category with the highest confidence in the commodity classification system predicted by the neural network model. By constraining the search object to the pre-judged category, the user can obtain a more accurate and pure commodity recommendation list, making the search recommendation strategy of the online store more reasonable, thereby comprehensively improving the economic benefits of the online store.

[0156] Please refer to Figure 10, A product search category recognition device provided to meet one of the purposes of the present application is a functional embodiment of the product search category recognition method of the present application. The device includes a request receiving module 1100, a coding construction module 1200, a feature integration module 1300, and a category recognition module 1400, where: The request receiving module 1100 is configured to receive a product search request submitted by a user to an online store and obtain the query term carried by the request; The coding construction module 1200 is configured to construct combined coding information, and the combined coding information includes the literal feature information of the query term, the personal feature information of the user, the store feature information of the online store, and the augmented feature information of the query term; The feature integration module 1300 is configured to construct the literal feature information and the fused feature information obtained by deep feature interaction of the combined coding information into comprehensive feature information; The category recognition module 1400 is configured to classify according to the deep semantic information of the comprehensive feature information and obtain the category in which the query term classification is mapped to the product classification system.

[0157] In some specific embodiments, the coding construction module 1200 includes: A literal coding unit configured to perform word embedding on the query term to vectorize and code it into literal feature information; An augmented coding unit configured to match an augmented vocabulary set composed of the associated words of the query term from a preset word list and vectorize and code the augmented vocabulary set into augmented feature information, where the associated words include synonyms and / or co-occurring words of the query term; A user coding unit configured to call multiple feature data of the user and vectorize and code the feature data into personal feature information; A store coding unit configured to call multiple feature data of the online store and vectorize and code the feature data into store feature information; A combined processing unit configured to perform multi-channel splicing on the literal feature information, augmented feature information, personal feature information, and store feature information to construct combined coding information.

[0158] In some specific embodiments, the augmented coding unit includes: A synonym matching subunit configured to semantically match the synonyms of the query term from a preset synonym list as its associated words; A co-occurrence matching subunit configured to semantically match the co-occurring words of the query term from a preset co-occurrence word list as its associated words; A set construction subunit configured to construct the synonyms and co-occurring words into an augmented vocabulary set of the query term; A vector conversion subunit configured to perform corresponding conversion on the associated words in the augmented vocabulary set according to a preset coding word list to achieve vectorization and obtain the augmented feature information of the query term.

[0159] In some deepened embodiments, the feature integration module 1300 includes: a feature interaction unit configured to perform deep feature interaction on the joint encoded information by using an attention layer to obtain fused feature information; and a splicing integration unit configured to perform multi-channel splicing on the fused feature information and the literal feature information of the query term by using a splicing layer to obtain integrated feature information.

[0160] In some deepened embodiments, the category recognition module 1400 includes: a feature extraction unit configured to extract deep semantic information from the integrated feature information by using a text feature extraction model; a classification mapping unit configured to map the deep semantic information to a classification space corresponding to a preset product classification system by using a classifier to obtain the confidence corresponding to each classification therein; and a category determination unit configured to determine the corresponding category in the product classification system for the query term according to the classification with the highest confidence.

[0161] In some extended embodiments, subsequent to the category recognition module 1400, it includes: a full-category retrieval module configured to retrieve product data whose product information matches the query term from the product database of the online store to form a product candidate list; a category ranking module configured to rank the product candidate list by category according to the confidence obtained for each classification in the classifier classification space to obtain a product recommendation list; and a list push module configured to push the product recommendation list to the user to respond to the product search request.

[0162] In some extended embodiments, subsequent to the category recognition module 1400, it includes: a category retrieval module configured to retrieve product data whose product information matches the query term and belongs to the category with the highest confidence from the product database of the online store to form a product recommendation list; and a list push module configured to push the product recommendation list to the user to respond to the product search request.

[0163] To solve the above technical problems, the embodiments of the present application further provide a computer device. As Figure 11As shown, it is a schematic diagram of the internal structure of a computer device. The computer device includes a processor, a computer-readable storage medium, a memory, and a network interface connected through a system bus. Among them, the computer-readable storage medium of the computer device stores an operating system, a database, and computer-readable instructions. The database can store a control information sequence. When the computer-readable instructions are executed by the processor, the processor can implement a method for identifying product search categories. The processor of the computer device is used to provide computing and control capabilities to support the operation of the entire computer device. The memory of the computer device can store computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor can execute the method for identifying product search categories of the present application. The network interface of the computer device is used to connect and communicate with a terminal. Those skilled in the art can understand, Figure 11 The structure shown in is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0164] In this embodiment, the processor is used to execute Figure 10 the specific functions of each module and its sub-modules in . The memory stores the program code and various types of data required to execute the above modules or sub-modules. The network interface is used for data transmission between the user terminal and the server. The memory in this embodiment stores the program code and data required to execute all modules / sub-modules in the product search category recognition device of the present application. The server can call the program code and data of the server to execute the functions of all sub-modules.

[0165] The present application also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors are caused to execute the steps of the method for identifying product search categories according to any embodiment of the present application.

[0166] The present application also provides a computer program product, including computer programs / instructions. When the computer programs / instructions are executed by one or more processors, the steps of the method according to any embodiment of the present application are implemented.

[0167] Those of ordinary skill in the art can understand that all or part of the processes of implementing the above embodiments of the present application can be completed by instructing relevant hardware through a computer program. This computer program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above embodiments of each method. Among them, the aforementioned storage medium can be a computer-readable storage medium such as a magnetic disk, an optical disc, a read-only memory (ROM), or a random access memory (RAM), etc.

[0168] In summary, the present application makes full use of the related words of the query term and the characteristics of the user and the online store in the process of determining the commodity category. Therefore, in the process of classification prediction, the reference information is richer, and more accurate prediction results can be obtained, enabling accurate prediction of the user's search intention, and thus matching a list of commodities that conform to the user's search intention for the user.

[0169] Those skilled in the art of the present technology can understand that the steps, measures, and solutions in the various operations, methods, and processes discussed in the present application can be alternated, changed, combined, or deleted. Further, the other steps, measures, and solutions in the various operations, methods, and processes discussed in the present application can also be alternated, changed, rearranged, decomposed, combined, or deleted. Further, the steps, measures, and solutions in the prior art that are the same as those disclosed in the present application can also be alternated, changed, rearranged, decomposed, combined, or deleted.

[0170] The above are only some embodiments of the present application. It should be noted that for those of ordinary skill in the art of the present technology, without departing from the principle of the present application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. A method for identifying product search categories, characterized in that, It includes the following steps: Receive a product search request submitted by a user to an online store, and obtain the query term carried by the request; Construct combined coding information, which includes the literal feature information of the query term, the personal feature information of the user, the store feature information of the online store, and the augmented feature information of the query term; Construct the combined feature information by combining the literal feature information with the fusion feature information obtained through deep feature interaction of the combined coding information, including: performing deep feature interaction on the combined coding information using an attention layer to obtain the fusion feature information; using a splicing layer to perform multi-channel splicing on the fusion feature information and the literal feature information of the query term to obtain the combined feature information; Classify according to the deep semantic information of the combined feature information to obtain the category in the product classification system to which the query term classification is mapped, including: using a text feature extraction model to extract the deep semantic information from the combined feature information; using a classifier to map the deep semantic information to the corresponding classification space of a preset product classification system to obtain the confidence corresponding to each classification; determining the corresponding category in the product classification system for the query term according to the classification with the highest confidence.

2. The method for identifying a product search category according to claim 1, wherein Construct the combined coding information, including the following steps: Perform word embedding on the query term to vectorize and encode it into literal feature information; Match the augmented vocabulary set composed of the related words of the query term from a preset word list, and vectorize and encode the augmented vocabulary set into augmented feature information, where the related words include the synonyms and / or co-occurring words of the query term; Call multiple feature data of the user, and vectorize and encode the feature data into personal feature information; Call multiple feature data of the online store, and vectorize and encode the feature data into store feature information; Perform multi-channel splicing on the literal feature information, augmented feature information, personal feature information, and store feature information to construct combined coding information.

3. The method for identifying a product search category according to claim 2, wherein, Match the augmented vocabulary set composed of the related words of the query term from a preset word list, and vectorize and encode the augmented vocabulary set into augmented feature information, including the following steps: Semantically match the synonyms of the query term from a preset synonym list as its related words; Semantically match the co-occurring words of the query term from a preset co-occurring word list as its related words; Construct the augmented vocabulary set of the query term from the synonyms and co-occurring words; According to a preset coding word list, perform corresponding conversion of the related words in the augmented vocabulary set into codes to achieve vectorization, and obtain the augmented feature information of the query term.

4. The method for identifying a product search category according to claim 1, wherein After the step of classifying according to the deep semantic information of the combined feature information to obtain the category in the product classification system to which the query term classification is mapped, it includes the following steps: Retrieve product data whose product information matches the query term from the product database of this online store to form a product candidate list; Sort the product candidate list by category according to the confidence obtained for each classification in the classifier classification space to obtain a product recommendation list; Push the product recommendation list to the user to respond to the product search request.

5. The method for identifying a product search category according to any one of claims 1 to 4, characterized in that, After the step of classifying according to the deep semantic information of the comprehensive feature information to obtain the category in the product classification system to which the query word is classified, the following steps are included: Retrieve product data from the product database of the online store, where the product information matches the query word and belongs to the category with the highest confidence, to form a product recommendation list; Push the product recommendation list to the user to respond to the product search request.

6. A commodity search category recognition device, characterized in that, Including: A request receiving module, configured to receive a product search request submitted by a user to an online store and obtain the query word carried by the request; An encoding construction module, configured to construct joint encoding information, where the joint encoding information includes the literal feature information of the query word, the personal feature information of the user, the store feature information of the online store, and the augmented feature information of the query word; A feature integration module, configured to construct the literal feature information and the fused feature information obtained by deep feature interaction of the joint encoding information into comprehensive feature information, including: performing deep feature interaction on the joint encoding information by using an attention layer to obtain fused feature information; using a splicing layer to perform multi-channel splicing on the fused feature information and the literal feature information of the query word to obtain comprehensive feature information; A category recognition module, configured to classify according to the deep semantic information of the comprehensive feature information to obtain the category in the product classification system to which the query word is classified, including: extracting deep semantic information from the comprehensive feature information by using a text feature extraction model; using a classifier to map the deep semantic information to the corresponding classification space of a preset product classification system to obtain the confidence corresponding to each classification; determining the corresponding category in the product classification system for the query word according to the classification with the highest confidence.

7. The commodity search category recognition device according to claim 6, wherein, The encoding construction module includes: A literal encoding unit, configured to perform word embedding on the query word to vectorize and encode it into literal feature information; An augmented encoding unit, configured to match the augmented vocabulary set composed of the related words of the query word from a preset word list, and vectorize and encode the augmented vocabulary set into augmented feature information, where the related words include synonyms and / or co-occurrence words of the query word; A user encoding unit, configured to call multiple feature data of the user and vectorize and encode the feature data into personal feature information; A store encoding unit, configured to call multiple feature data of the online store and vectorize and encode the feature data into store feature information; A joint processing unit, configured to perform multi-channel splicing on the literal feature information, augmented feature information, personal feature information, and store feature information to construct joint encoding information.

8. The merchandise search category recognition device according to claim 7, characterized in that, The augmented encoding unit includes: A synonym matching subunit, configured to semantically match the synonyms of the query word from a preset synonym list as its related words; A co-occurrence matching subunit, configured to semantically match the co-occurrence words of the query word from a preset co-occurrence word list as its related words; A set construction subunit for constructing the synonyms and co-occurrence words into an augmented vocabulary set of the query term; A vector conversion subunit for correspondingly converting the related words in the augmented vocabulary set into encodings to achieve vectorization according to a preset encoding vocabulary table, and obtaining augmented feature information of the query term.

9. A computer device, comprising a central processing unit and a memory, characterized in that, The central processing unit is configured to call and run a computer program stored in the memory to execute the steps of the method according to any one of claims 1 to 5.

10. A computer-readable storage medium, characterized in that, It stores, in the form of computer-readable instructions, a computer program implemented according to the method according to any one of claims 1 to 5. When the computer program is called and run by a computer, it executes the steps included in the corresponding method.

Images