Commodity search method and device

Abstract

The present application relates to the field of commodity retrieval technology, and in particular to a commodity retrieval method and device, the method comprising obtaining user information and a current search term from a user terminal; predicting the user's intent based on the user information in combination with a preset intent recognition model to obtain a prediction result; determining the part of speech, search scope, and initial weight of the current search term based on the prediction result; searching for recommended commodities within the search scope based on the current search term; calculating the rating of each recommended commodity based on the initial weight; and sorting each recommended commodity in descending order of rating value and pushing the result to the user terminal. The present application can achieve the purpose of recommending commodities required by users more quickly, which is conducive to improving the user's browsing experience and increasing the order rate of commodities.

Description

Technical Field

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

[0002] With the development of e-commerce, product retrieval technology has attracted more and more attention.

[0003] For industrial commodities that are not production raw materials, the number of categories, manufacturers, suppliers, and even brand names and product names under the same category is huge. Existing retrieval methods generally rely on search engines to index the entire database based on the current search terms entered by the user and recommend products that meet the current search terms to the user. However, the recommended results are often difficult to meet the user's actual purchasing needs. At the same time, it takes a long time to retrieve recommended products from the huge number of industrial commodities, making it difficult for search engines to quickly recommend products that users need.

[0004] With respect to the above-mentioned related technologies, the inventors have discovered that the existing search methods for industrial products have the problem of being difficult to quickly recommend products that users need. Summary of the Invention

[0005] In order to find the products that users need more quickly, this application provides a product retrieval method and device.

[0006] In a first aspect, the present application provides a product search method.

[0007] This application is achieved through the following technical solutions:

[0008] A product search method includes the following steps:

[0009] Get user information and current search terms from the user side;

[0010] Based on the user information, combined with a preset intention recognition model, the user intention is predicted to obtain a prediction result;

[0011] Determining the part of speech, search scope, and initial weight of the current search term based on the prediction result;

[0012] Based on the current search term, search within the search range to obtain recommended products;

[0013] Calculating the score of each of the recommended products based on the initial weights;

[0014] The recommended products are sorted in descending order according to their ratings and then pushed to the user end.

[0015] In a preferred example, the present application can be further configured as follows: the training step of the intent recognition model includes:

[0016] Obtaining user information as training corpus, wherein the user information includes the material name of the product, the user's historical search terms and the associated information formed by browsing product information, and the user's quotation data information;

[0017] Aggregating, removing duplicates, and correcting spelling errors on the training corpus in sequence to obtain a target dataset;

[0018] Inputting a preset proportion of the target data set into a neural network model for training and outputting user intent;

[0019] When the accuracy of the neural network model reaches a preset value, the neural network model with the accuracy reaching the preset value is output as the intention recognition model.

[0020] In a preferred example, the present application may be further configured as follows: the step of determining the part of speech, search scope and initial weight of the current search term according to the prediction result includes:

[0021] According to the prediction result, matching the historical search terms corresponding to the current search term from a preset user intent table, wherein the user intent table includes the user's historical search terms, corresponding parts of speech, initial weights corresponding to the parts of speech, and search scopes corresponding to the parts of speech;

[0022] Based on the parts of speech corresponding to the historical search terms, the part of speech with the largest initial weight is selected as the part of speech of the current search term, and the initial weight corresponding to the part of speech is obtained, and the search scope corresponding to the part of speech with the largest initial weight is selected as the search scope of the current search term.

[0023] In a preferred example, the present application may be further configured as follows: based on the initial weight, the step of calculating the score value of each of the recommended products includes:

[0024] Obtain at least two product rating items, as well as initial scores and item weights corresponding to each item, wherein the product rating items include part-of-speech items;

[0025] Using the initial weight as the item weight of the corresponding part-of-speech item of the corresponding recommended product;

[0026] The rating value of the recommended product is calculated based on the initial score and item weight corresponding to the product rating item.

[0027] In a preferred example, the present application can be further configured as follows:

[0028] The product rating items also include product name items, brand items and attribute items. Let the initial score of the product name item * product weight + the initial score of the brand item * brand weight + the initial score of the attribute item * attribute weight + the initial score of the part of speech item corresponding to the initial weight * part of speech weight to obtain the rating value of the recommended product, wherein the attribute weight < the product weight < the initial weight < brand weight, and the attribute weight + the product weight + the brand weight = 1.

[0029] In a preferred example, the present application can be further configured as follows:

[0030] Obtain the user's browsing product information and browsing time in real time;

[0031] Based on the browsed product information and the browsing time, the initial weight of the current search term is updated in real time.

[0032] In a preferred example, the present application can be further configured as follows: based on the browsed product information and the browsed time, the step of updating the initial weight of the current search term in real time includes:

[0033] If the browsing time reaches a preset value and there is no browsing product information on the user side, the initial weights of the current search terms with a preset number of top rankings are downgraded to obtain a real-time weight;

[0034] The real-time weight is used to replace the corresponding initial weight of the current search term.

[0035] In a preferred example, the present application can be further configured as follows:

[0036] Regularly collect statistics on the browsed products of each user terminal;

[0037] Calculate the click-through rate of the same product based on the product information browsed by each user terminal;

[0038] Based on the click-through rate of the product, the initial weight of the current search term is updated.

[0039] In a preferred example, the present application can be further configured as follows: the step of updating the initial weight of the current search term based on the click-through rate of the product includes:

[0040] According to the click-through rate of the product, the initial weight of the current search term is increased to obtain a changed weight, wherein the greater the click-through rate of the product, the greater the initial weight of the current search term;

[0041] The changed weight is used to update the initial weight of the current search term.

[0042] In a second aspect, the present application provides a product search device.

[0043] This application is achieved through the following technical solutions:

[0044] A commodity search device, comprising:

[0045] Data module, used to obtain user information and current search terms from the user end;

[0046] An intention prediction module is used to predict the user's intention based on the user information and a preset intention recognition model to obtain a prediction result;

[0047] An adjustment module, configured to determine the part of speech, search scope, and initial weight of the current search term based on the prediction result;

[0048] A search module, configured to search for recommended products within the search range based on the current search term;

[0049] A scoring module, configured to calculate a scoring value for each of the recommended products based on the initial weights;

[0050] The push module is used to sort the recommended products in descending order of their ratings and then push them to the user end.

[0051] In a preferred example, the present application can be further configured as follows: the intention prediction module includes:

[0052] A training corpus unit is used to obtain user information as training corpus, wherein the user information includes the material name of the product, the user's historical search terms and the associated information formed by browsing product information, and the user's quotation data information;

[0053] A target data set unit is used to aggregate, remove duplicates and correct spelling errors on the training corpus in sequence to obtain a target data set;

[0054] A training unit, configured to input a preset proportion of the target data set into a neural network model for training and output user intent;

[0055] The intention recognition model unit is used to output the neural network model with the accuracy reaching the preset value as the intention recognition model when the accuracy of the neural network model reaches the preset value.

[0056] In a preferred example, the present application can be further configured as follows: the adjustment module includes:

[0057] a matching unit, configured to match, based on the prediction result, a historical search term corresponding to the current search term from a preset user intent table, wherein the user intent table includes the user's historical search terms, corresponding parts of speech, initial weights corresponding to the parts of speech, and search scopes corresponding to the parts of speech;

[0058] The adjustment unit is used to select the part of speech with the largest initial weight as the part of speech of the current search term based on the part of speech corresponding to the historical search term, obtain the initial weight corresponding to the part of speech, and select the search range corresponding to the part of speech with the largest initial weight as the search range of the current search term.

[0059] In a preferred example, the present application can be further configured as follows: the scoring module includes:

[0060] A scoring unit, configured to obtain at least two product rating items, and initial scores and item weights corresponding to each item, wherein the product rating items include part-of-speech items;

[0061] an optimization unit, configured to use the initial weight as the item weight of the corresponding part-of-speech item of the corresponding recommended product;

[0062] A calculation unit is used to calculate the rating value of the recommended product based on the initial score and item weight corresponding to the product rating item.

[0063] In a preferred example, the present application can be further configured as follows: the product rating items of the rating unit further include product name items, brand items, and attribute items, and the calculation unit further includes:

[0064] The calculation subunit is used to obtain the rating value of the recommended product by adding the initial score of the product name item * the product weight + the initial score of the brand item * the brand weight + the initial score of the attribute item * the attribute weight + the initial score of the part-of-speech item corresponding to the initial weight * the part-of-speech weight, wherein the attribute weight < the product weight < the initial weight < the brand weight, and the attribute weight + the product weight + the brand weight = 1.

[0065] In a preferred example, the present application can be further configured as follows:

[0066] Real-time detection module, used to obtain the user's browsing product information and browsing time in real time;

[0067] The initial weight real-time updating module is used to update the initial weight of the current search term in real time based on the browsed product information and the browsing time.

[0068] In a preferred example, the present application can be further configured as follows: the initial weight real-time update module includes:

[0069] a weight reduction unit, configured to reduce the initial weights of a preset number of the current search terms ranked at the top to obtain real-time weights when the browsing time reaches a preset value and there is no browsing product information on the user terminal;

[0070] A real-time updating unit is used to replace the initial weight of the corresponding current search term with the real-time weight.

[0071] In a preferred example, the present application can be further configured as follows:

[0072] Statistics module, used to regularly collect statistics on the browsed products of each user terminal;

[0073] The click-through rate module is used to calculate the click-through rate of the same product based on the product information browsed by each user terminal;

[0074] The initial weight periodic updating module updates the initial weight of the current search term based on the click-through rate of the product.

[0075] In a preferred example, the present application can be further configured as follows: the initial weight periodic update module includes:

[0076] a weighting unit for increasing the initial weight of the current search term according to the click-through rate of the product to obtain a changed weight, wherein the greater the click-through rate of the product, the greater the initial weight of the current search term;

[0077] The initial weight periodic updating unit is used to update the initial weight of the current search term using the changed weight.

[0078] In a third aspect, the present application provides a computer device.

[0079] This application is achieved through the following technical solutions:

[0080] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any one of the above-mentioned commodity retrieval methods when executing the computer program.

[0081] In a fourth aspect, the present application provides a computer-readable storage medium.

[0082] This application is achieved through the following technical solutions:

[0083] A computer-readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the steps of any one of the above-mentioned product retrieval methods.

[0084] In summary, compared with the prior art, the technical solution provided by this application has at least the following beneficial effects:

[0085] Obtain user information and current search terms from the user end; predict user intent based on user information and combined with a preset intent recognition model to obtain a prediction result; determine the part of speech, search scope and initial weight of the current search term based on the prediction result; based on the current search term, search within the determined search scope to obtain recommended products, without the need for a global, whole-site search, resulting in higher retrieval efficiency; at the same time, calculate the score value of each recommended product based on the initial weight, sort each recommended product in descending order according to the score value, and push it to the user end to adjust the order of recommended products according to user intent, and give priority to recommending products that meet user intent; it can recommend products that meet user needs more quickly, and the retrieved products are more in line with user needs, which is conducive to improving the user's browsing experience and thereby increasing the order rate of products. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] Figure 1 A schematic diagram of the main flow of a product retrieval method provided as an exemplary embodiment of the present application.

[0087] Figure 2 A flowchart of training an intent recognition model for a product retrieval method is provided as another exemplary embodiment of the present application.

[0088] Figure 3 A flowchart of a product retrieval method for determining the part of speech, retrieval scope, and initial weight of a current search term based on prediction results is provided as another exemplary embodiment of the present application.

[0089] Figure 4 A flowchart of a product retrieval method for calculating the score value of each recommended product based on the initial weight is provided as an exemplary embodiment of the present application.

[0090] Figure 5 A flowchart of a product retrieval method for updating the initial weight of a current search term in real time is provided as an exemplary embodiment of the present application.

[0091] Figure 6 A flowchart of a product retrieval method for periodically updating the initial weight of a current search term is provided as an exemplary embodiment of the present application.

[0092] Figure 7 A structural block diagram of a product search device provided as an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0093] This specific embodiment is merely an explanation of the present application and is not a limitation of the present application. After reading this specification, those skilled in the art may make non-creative modifications to the present embodiment as needed, but as long as they are within the scope of the claims of the present application, they are protected by the patent law.

[0094] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0095] In this document, the term "and / or" simply describes a relationship between related objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " in this document, unless otherwise specified, generally indicates an "or" relationship between the related objects.

[0096] The embodiments of the present application are described in further detail below with reference to the accompanying drawings.

[0097] Reference Figure 1 , an embodiment of the present application provides a product retrieval method, and the main steps of the method are described as follows.

[0098] S1: Obtain user information and current search terms from the user end;

[0099] S2: Based on the user information, combined with a preset intention recognition model, predict the user intention to obtain a prediction result;

[0100] S3: Determine the part of speech, search scope, and initial weight of the current search term based on the prediction result;

[0101] S4: Based on the current search term, search within the search range to obtain recommended products;

[0102] S5: Calculating the score of each of the recommended products based on the initial weights;

[0103] S6: Sort the recommended products in descending order of their ratings and push them to the user end.

[0104] Reference Figure 2 In one embodiment, the training step of the intent recognition model includes:

[0105] S21: Obtain user information as training corpus, wherein the user information includes the material name of the product, the user's historical search terms and the associated information formed by browsing product information, and the user's quotation data information;

[0106] S22: Aggregate, remove duplicates, and correct spelling errors on the training corpus in sequence to obtain a target dataset;

[0107] S23: Inputting a preset proportion of the target data set into a neural network model for training, and outputting the user intention;

[0108] S24: When the accuracy of the neural network model reaches a preset value, the neural network model with the accuracy reaching the preset value is output as the intention recognition model.

[0109] Reference Figure 3 In one embodiment, S3: determining the part of speech, search scope and initial weight of the current search term based on the prediction result includes:

[0110] S31: According to the prediction result, matching the historical search terms corresponding to the current search term from a preset user intent table, wherein the user intent table includes the user's historical search terms, corresponding parts of speech, initial weights corresponding to the parts of speech, and search scopes corresponding to the parts of speech;

[0111] S32: Based on the parts of speech corresponding to the historical search terms, select the part of speech with the largest initial weight as the part of speech of the current search term, obtain the initial weight corresponding to the part of speech, and select the search scope corresponding to the part of speech with the largest initial weight as the search scope of the current search term.

[0112] Reference Figure 4 In one embodiment, S5: Based on the initial weights, the step of calculating the score of each of the recommended products includes:

[0113] S51: Obtain at least two product rating items, and initial scores and item weights corresponding to each item, wherein the product rating items include part-of-speech items;

[0114] S52: Using the initial weight as the item weight of the corresponding part-of-speech item of the corresponding recommended product;

[0115] S53: Calculate the rating value of the recommended product based on the initial score and item weight corresponding to the product rating item.

[0116] In one embodiment, the following steps are also included:

[0117] S54: The product rating items also include product name items, brand items and attribute items. Let the initial score of the product name item * product weight + the initial score of the brand item * brand weight + the initial score of the attribute item * attribute weight + the initial score of the part of speech item corresponding to the initial weight * part of speech weight to obtain the rating value of the recommended product, wherein the attribute weight < the product weight < the initial weight < brand weight, and the attribute weight + the product weight + the brand weight = 1.

[0118] Reference Figure 5 In one embodiment, the following steps are also included:

[0119] S71: Obtaining the browsing product information and browsing time of the user terminal in real time;

[0120] Based on the browsed product information and the browsing time, the initial weight of the current search term is updated in real time.

[0121] In one embodiment, the step of updating the initial weight of the current search term in real time based on the browsed product information and the browsing time includes:

[0122] S811: If the browsing time reaches a preset value and there is no browsing product information on the user terminal, the initial weights of the current search terms ranked at the top of a preset number are downgraded to obtain a real-time weight;

[0123] S812: Use the real-time weight to replace the corresponding initial weight of the current search term.

[0124] Reference Figure 6 In one embodiment, the following steps are also included:

[0125] S721: Regularly collect statistics on the browsed products of each user terminal;

[0126] S722: Calculate the click rate of the same product based on the product information browsed by each user terminal;

[0127] Based on the click-through rate of the product, the initial weight of the current search term is updated.

[0128] In one embodiment, the step of updating the initial weight of the current search term based on the click-through rate of the product includes:

[0129] S7231: According to the click-through rate of the product, the initial weight of the current search term is increased to obtain a changed weight, wherein the greater the click-through rate of the product, the greater the initial weight of the current search term;

[0130] S7232: Use the changed weight to update the initial weight of the current search term.

[0131] The above embodiments are described in detail as follows.

[0132] Specifically, after logging into the user terminal, the user enters one, two or more keywords in the search engine.

[0133] The backend obtains the user information of the user end and uses one, two or more keywords entered as the current search terms.

[0134] Based on user information, combined with the preset intention recognition model, the user intention is predicted to obtain the prediction result, thereby achieving the purpose of identifying user intention based on user historical data.

[0135] Through the user account information in the user information, the corresponding historical browsing data is indexed, and the preset intention recognition model is combined to predict the user intention, and the association information between historical keywords and clicked products, that is, the prediction result, is obtained to realize the prediction of user intention.

[0136] If no results are found when indexing the historical browsing data corresponding to the user account information in the user information, the historical browsing data corresponding to similar users is indexed, and the user intent is predicted based on the preset intent recognition model to obtain the association information between historical keywords and clicked products.

[0137] Similar users are defined as users whose gender is consistent with the user information of the user terminal and whose age is different from that of the user by a preset value. In this embodiment, the preset value may be 5.

[0138] The intent recognition model uses a multilayer perceptron neural network model. This model consists of multiple layers of neurons. The input of each layer is the output of the previous layer, and the output of this layer is the input of the next layer, connected in sequence. In this embodiment, a three-layer perceptron neural network is used, including an input layer, a hidden layer, and an output layer.

[0139] When training the intent recognition model, first obtain user information as training corpus.

[0140] User information includes the material name of the product, the user's historical search terms and the associated information formed by browsing product information, and the user's quotation data information.

[0141] The associated information formed by material names, user historical search terms and browsed product information can be obtained through the historical database, and user quotation data information can be obtained by manual marking.

[0142] The training corpus is aggregated, deduplicated, and spell-corrected to produce the target dataset. Aggregation involves combining synonyms or near-synonyms to create a vocabulary summary. Deduplication involves removing duplicate words by retaining only the first occurrence. Spelling correction corrects misspelled words and standardizes the word format. After preprocessing the training corpus, the vocabulary is partitioned using a database-tree structure to produce the target dataset.

[0143] By selecting the product material name, user search terms and the associated information formed by browsing product information, and user quotation data from a large amount of search data as the training corpus of the model, and pre-processing the training corpus by aggregation, deduplication and spelling correction, a target data set is formed for machine learning, achieving the purpose of using the messy and numerous search data for machine learning, which is conducive to the intelligent recognition of user intent.

[0144] Furthermore, the target dataset is optimized by deleting ambiguous or incorrect training corpus to remove the impact of dirty data in the search results on model training, improve the quality of the training corpus, and help improve the accuracy of the model's output results.

[0145] The target data set of a preset ratio is input into the neural network model for training to output the user intention. In this embodiment, the target data set is divided into a ratio of 2:8 to obtain a verification data set and a training data set.

[0146] The training data set is input into the neural network model for training.

[0147] The validation dataset meets the preset design rules, such as "non-punctuation marks", "no Chinese characters", "elimination of function words", etc., to verify the training results of the model.

[0148] When the training accuracy of the neural network model reaches a preset value, the neural network model with the preset accuracy is output as the intent recognition model. In this embodiment, when the accuracy of the neural network model reaches 80%, the neural network model is output as the intent recognition model.

[0149] Based on historical users and historical browsing data, a user intent table is created. The user intent table includes the user's historical search terms, corresponding parts of speech, initial weights corresponding to the parts of speech, and search scopes corresponding to the parts of speech.

[0150] Based on the prediction results, match the historical search terms corresponding to the current search term from the preset user intent table;

[0151] Based on the parts of speech corresponding to the historical search terms, the part of speech with the largest initial weight is selected as the part of speech of the current search term, and the initial weight corresponding to the part of speech is obtained.

[0152] The search scope corresponding to the part of speech with the largest initial weight is selected as the search scope of the current search term.

[0153] Based on the current search term, search within the search scope to obtain recommended products.

[0154] For example, if a user enters the keyword "apple," the corresponding parts of speech are product type and product brand. Furthermore, if the user's historical search data indicates that they have always searched for mobile phones, the initial weights for product type and brand are set to 0.3 and 0.7, respectively. Because the initial weight for brand is greater than the initial weight for product type, the corresponding part of speech for "apple" is determined to be "product brand," and the search scope for "product brand" is determined to be the mobile phone category.

[0155] Based on the initial weights, calculate the ratings for each recommended product.

[0156] Specifically, product rating items include product name, part of speech, brand, and attribute items. Initial scores and item weights are pre-set for each of these items. These initial weights are used as the item weights for the corresponding part of speech items of the recommended products. The sum of the item weights for each product rating item is 1.

[0157] At least two product rating items, as well as initial scores and item weights corresponding to each item, are obtained, wherein the product rating items include part-of-speech items.

[0158] The rating value of the recommended product is calculated based on the initial score and item weight corresponding to the product rating item.

[0159] In one embodiment, the product name item, part of speech item, brand item, and attribute item, as well as the initial score and item weight corresponding to each product rating item are obtained.

[0160] Let the initial score of the product name item * product weight + the initial score of the brand item * brand weight + the initial score of the attribute item * attribute weight + the initial score of the part of speech item corresponding to the initial weight * part of speech weight to obtain the rating value of the recommended product, where the attribute weight < the product weight < the initial weight < brand weight, and the attribute weight + the product weight + the brand weight = 1.

[0161] For example, the current search term is gloves. By counting the historical browsing data of historical users, we get a click probability of 0.4 for protective gloves and a click probability of 0.6 for disposable gloves. Among them, the initial score of disposable gloves is 60 and the initial score of protective gloves is 50. Therefore, the part of speech of gloves is determined to be disposable gloves, and the initial weight is 0.6.

[0162] When ranking the search results, when calculating the score value of any search result corresponding to gloves, first obtain the product rating items corresponding to the search results one by one and their corresponding initial scores and item weights, as shown in Table 1 below.

[0163] Table 1

[0164] Product rating items Initial score Item Weight Product Name 30 0.2 brand 150 0.7 property 10 0.1 disposable gloves 60 0.6

[0165] The score value of the search result is 30*0.2+150*0.7+10*0.1+60*0.6=148.

[0166] Calculate the score of all search results corresponding to gloves according to the above method, and use it as the score of each recommended product.

[0167] The recommended products are sorted in descending order according to their ratings and then pushed to the user end.

[0168] By identifying the user's intent, the part of speech and search scope of each keyword entered by the user are determined. Different parts of speech have different initial weights.

[0169] The initial weight is related to the search results. The larger the initial weight, the higher the product rating.

[0170] In order to make the initial weight of the product more accurate, we further obtain the user's browsing product information and browsing time in real time;

[0171] Based on the browsed product information and the browsing time, the initial weight of the current search term is updated in real time, including:

[0172] If the browsing time reaches a preset value and there is no browsing product information on the user side, the initial weights of the current search terms with a preset number of top rankings are downgraded to obtain a real-time weight;

[0173] The real-time weight is used to replace the corresponding initial weight of the current search term.

[0174] For example, if the user does not click to view any of the products on the first page of search results within a preset time range and moves on to browse the second page of products, the initial weights of all products on the first page will be downgraded in real time to adjust the initial weights of the products browsed by the user in real time, so that the product search results are relevant to the user intent recognition results, and products that meet user needs can be recommended, making the recommendation results more accurate and helping to increase the order rate of products.

[0175] Alternatively, the initial weight can be updated in real time based on the real-time feedback results of the current user, such as click-through rate, browsing time, transaction rate, etc.

[0176] In order to make the initial weight of the product more accurate, in one embodiment, the browsing product information of each user terminal is regularly counted;

[0177] Calculate the click-through rate of the same product based on the product information browsed by each user terminal;

[0178] Based on the click-through rate of the product, the initial weight of the current search term is updated, including:

[0179] According to the click-through rate of the product, the initial weight of the current search term is increased to obtain a changed weight, wherein the greater the click-through rate of the product, the greater the initial weight of the current search term;

[0180] The changed weight is used to update the initial weight of the current search term.

[0181] By regularly collecting historical search data of all users and uniformly adjusting the initial weights of products, the adjustment results of the initial weights of products are made more representative. The set initial weights can better meet the search requirements, making the recommendation results more accurate and helping to increase the order rate of products.

[0182] Furthermore, in order to optimize query performance, when searching for the current search term, query performance tuning is implemented based on ElasticSearch technology.

[0183] Specifically, multi-level cache optimization is set. In this embodiment, two layers of cache optimization are set. Among them, the first layer is the Redis cache (central cache) on the cloud server, and the second layer is the local cache (local node cache) used by the program itself. During retrieval, the product data queried for the first time can be loaded from the main database into the first layer cache, and then from the first layer cache into the second layer cache, so that when the same product is queried for the second time, the product data can be obtained from the second layer cache. If there is no target product in the second layer cache, check whether it exists in the first layer cache. When there is no product data in both layers of cache, the target product data is obtained from the main database again. The retrieval speed is greatly improved.

[0184] Of course, you can also clear existing data in the cache to make room for new data. For example, you can set it to clear once a day, or clear cached data when the number of queries per unit time falls below a preset threshold.

[0185] When optimizing query performance using ElasticSearch technology, you can customize the search scope. For a specific type of product, only searches within the corresponding search scope. For example, if the keyword is "mobile phone," only data within the corresponding mobile phone range in the database will be searched.

[0186] In summary, a product retrieval method obtains user information and current search terms from the user end; based on the user information, the user intent is predicted in combination with a preset intent recognition model to obtain a prediction result; according to the prediction result, the part of speech, search scope and initial weight of the current search term are determined; based on the current search term, recommended products are searched within the determined search scope without the need for a global, whole-site search, which results in higher retrieval efficiency; at the same time, the score value of each recommended product is calculated based on the initial weight, and the recommended products are sorted in descending order according to the score value and pushed to the user end to adjust the order of recommended products according to the user intent, and give priority to recommending products that meet the user intent; products that meet user needs can be recommended more quickly, and the retrieved products are more in line with user needs, which is conducive to improving the user's browsing experience and thereby increasing the order rate of products.

[0187] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0188] Reference Figure 7 The present application also provides a product search device, which corresponds to the product search method in the above embodiment. The product search device includes:

[0189] Data module, used to obtain user information and current search terms from the user end;

[0190] An intention prediction module is used to predict the user's intention based on the user information and a preset intention recognition model to obtain a prediction result;

[0191] An adjustment module, configured to determine the part of speech, search scope, and initial weight of the current search term based on the prediction result;

[0192] A search module, configured to search for recommended products within the search range based on the current search term;

[0193] A scoring module, configured to calculate a scoring value for each of the recommended products based on the initial weights;

[0194] The push module is used to sort the recommended products in descending order of their ratings and then push them to the user end.

[0195] Furthermore, the intention prediction module includes,

[0196] A training corpus unit is used to obtain user information as training corpus, wherein the user information includes the material name of the product, the user's historical search terms and the associated information formed by browsing product information, and the user's quotation data information;

[0197] A target data set unit is used to aggregate, remove duplicates and correct spelling errors on the training corpus in sequence to obtain a target data set;

[0198] A training unit, configured to input a preset proportion of the target data set into a neural network model for training and output user intent;

[0199] The intention recognition model unit is used to output the neural network model with the accuracy reaching the preset value as the intention recognition model when the accuracy of the neural network model reaches the preset value.

[0200] Furthermore, the adjustment module includes,

[0201] a matching unit, configured to match, based on the prediction result, a historical search term corresponding to the current search term from a preset user intent table, wherein the user intent table includes the user's historical search terms, corresponding parts of speech, initial weights corresponding to the parts of speech, and search scopes corresponding to the parts of speech;

[0202] The adjustment unit is used to select the part of speech with the largest initial weight as the part of speech of the current search term based on the part of speech corresponding to the historical search term, obtain the initial weight corresponding to the part of speech, and select the search range corresponding to the part of speech with the largest initial weight as the search range of the current search term.

[0203] Furthermore, the scoring module includes,

[0204] A scoring unit, configured to obtain at least two product rating items, and initial scores and item weights corresponding to each item, wherein the product rating items include part-of-speech items;

[0205] an optimization unit, configured to use the initial weight as the item weight of the corresponding part-of-speech item of the corresponding recommended product;

[0206] A calculation unit is used to calculate the rating value of the recommended product based on the initial score and item weight corresponding to the product rating item.

[0207] Furthermore, the product rating items of the rating unit further include product name items, brand items and attribute items, and the calculation unit further includes:

[0208] The calculation subunit is used to obtain the rating value of the recommended product by adding the initial score of the product name item * the product weight + the initial score of the brand item * the brand weight + the initial score of the attribute item * the attribute weight + the initial score of the part-of-speech item corresponding to the initial weight * the part-of-speech weight, wherein the attribute weight < the product weight < the initial weight < the brand weight, and the attribute weight + the product weight + the initial weight + the brand weight = 1.

[0209] Furthermore, a product search device also includes:

[0210] Real-time detection module, used to obtain the user's browsing product information and browsing time in real time;

[0211] The initial weight real-time updating module is used to update the initial weight of the current search term in real time based on the browsed product information and the browsing time.

[0212] Furthermore, the initial weight real-time update module includes,

[0213] a weight reduction unit, configured to reduce the initial weights of a preset number of the current search terms ranked at the top to obtain real-time weights when the browsing time reaches a preset value and there is no browsing product information on the user terminal;

[0214] A real-time updating unit is used to replace the initial weight of the corresponding current search term with the real-time weight.

[0215] Furthermore, a product search device also includes:

[0216] Statistics module, used to regularly collect statistics on the browsed products of each user terminal;

[0217] The click-through rate module is used to calculate the click-through rate of the same product based on the product information browsed by each user terminal;

[0218] The initial weight periodic updating module updates the initial weight of the current search term based on the click-through rate of the product.

[0219] Furthermore, the initial weight periodic update module includes,

[0220] a weighting unit for increasing the initial weight of the current search term according to the click-through rate of the product to obtain a changed weight, wherein the greater the click-through rate of the product, the greater the initial weight of the current search term;

[0221] The initial weight periodic updating unit is used to update the initial weight of the current search term using the changed weight.

[0222] The specific definition of a product search device can be found in the definition of a product search method above and will not be repeated here. Each module in the above-mentioned product search device can be implemented in whole or in part through software, hardware, or a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in hardware form, or can be stored in a memory in a computer device in software form, so that the processor can call and execute the corresponding operations of each of the above modules.

[0223] In one embodiment, a computer device is provided, which may be a server. The computer device includes a processor, memory, a network interface, and a database connected via a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is configured to communicate with an external terminal via a network connection. When executed by the processor, the computer program implements any of the aforementioned product retrieval methods.

[0224] In one embodiment, a computer-readable storage medium is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are implemented:

[0225] S1: Obtain user information and current search terms from the user end;

[0226] S2: Based on the user information, combined with a preset intention recognition model, predict the user intention to obtain a prediction result;

[0227] S3: Determine the part of speech, search scope, and initial weight of the current search term based on the prediction result;

[0228] S4: Based on the current search term, search within the search range to obtain recommended products;

[0229] S5: Calculating the score of each of the recommended products based on the initial weights;

[0230] S6: Sort the recommended products in descending order of their ratings and push them to the user end.

[0231] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0232] Those skilled in the art will clearly understand that for the sake of convenience and brevity in description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

Claims

1. A product search method, characterized in that: The following steps are included: Get user information and current search terms from the user side; Based on the user information, the user intention is predicted in combination with the preset intention recognition model to obtain a prediction result. Determining the part of speech, search scope, and initial weight of the current search term based on the prediction result; Based on the current search term, search within the search range to obtain recommended products; Calculating the score of each of the recommended products based on the initial weights; Sort the recommended products in descending order of their ratings and push them to the user; The step of determining the part of speech, search scope, and initial weight of the current search term based on the prediction result further includes matching historical search terms corresponding to the current search term from a preset user intent table based on the prediction result, wherein the user intent table includes the user's historical search terms, corresponding parts of speech, initial weights corresponding to the parts of speech, and search scopes corresponding to the parts of speech; Based on the parts of speech corresponding to the historical search terms, the part of speech with the largest initial weight is selected as the part of speech of the current search term, and the initial weight corresponding to the part of speech is obtained, and the search scope corresponding to the part of speech with the largest initial weight is selected as the search scope of the current search term.

2. The product search method according to claim 1, characterized in that: The training steps of the intent recognition model include: Obtaining user information as training corpus, wherein the user information includes the material name of the product, the user's historical search terms and the associated information formed by browsing product information, and the user's quotation data information; Aggregating, removing duplicates, and correcting spelling errors on the training corpus in sequence to obtain a target dataset; Inputting a preset proportion of the target data set into a neural network model for training and outputting user intent; When the accuracy of the neural network model reaches a preset value, the neural network model with the accuracy reaching the preset value is output as the intention recognition model.

3. The product search method according to any one of claims 1 to 2, characterized in that: Based on the initial weights, the step of calculating the ratings of the recommended products includes: Obtain at least two product rating items, as well as initial scores and item weights corresponding to each item, wherein the product rating items include part-of-speech items; Using the initial weight as the item weight of the corresponding part-of-speech item of the corresponding recommended product; The rating value of the recommended product is calculated based on the initial score and item weight corresponding to the product rating item.

4. The product search method according to claim 3, characterized in that: Also includes the following steps, The product rating items also include product name items, brand items and attribute items. Let the initial score of the product name item * product weight + the initial score of the brand item * brand weight + the initial score of the attribute item * attribute weight + the initial score of the part of speech item corresponding to the initial weight * part of speech weight to obtain the rating value of the recommended product, wherein the attribute weight < the product weight < the initial weight < brand weight, and the attribute weight + the product weight + the brand weight = 1.

5. The product search method according to claim 3, wherein: Also includes the following steps, Obtain the user's browsing product information and browsing time in real time; Based on the browsed product information and the browsing time, the initial weight of the current search term is updated in real time.

6. The product search method according to claim 5, characterized in that: The step of updating the initial weight of the current search term in real time based on the browsed product information and the browsing time includes: If the browsing time reaches a preset value and there is no browsing product information on the user side, the initial weights of the current search terms with a preset number of top rankings are downgraded to obtain a real-time weight; The real-time weight is used to replace the corresponding initial weight of the current search term.

7. The product search method according to claim 3, characterized in that: Also includes the following steps, Regularly collect statistics on the browsed products of each user terminal; Calculate the click-through rate of the same product based on the product information browsed by each user terminal; Based on the click-through rate of the product, the initial weight of the current search term is updated.

8. The product search method according to claim 7, characterized in that: The step of updating the initial weight of the current search term based on the click-through rate of the product includes: According to the click-through rate of the product, the initial weight of the current search term is increased to obtain a changed weight, wherein the greater the click-through rate of the product, the greater the initial weight of the current search term; The changed weight is used to update the initial weight of the current search term.

9. A product search device, characterized in that: include, Data module, used to obtain user information and current search terms from the user end; An intention prediction module is used to predict the user's intention based on the user information and a preset intention recognition model to obtain a prediction result; An adjustment module, configured to determine the part of speech, search scope, and initial weight of the current search term based on the prediction result; a matching unit, configured to match, based on the prediction result, a historical search term corresponding to the current search term from a preset user intent table, wherein the user intent table includes the user's historical search terms, corresponding parts of speech, initial weights corresponding to the parts of speech, and search scopes corresponding to the parts of speech; an adjusting unit, configured to select the part of speech with the largest initial weight as the part of speech of the current search term based on the parts of speech corresponding to the historical search terms, obtain the initial weights corresponding to the parts of speech, and select the search scope corresponding to the part of speech with the largest initial weight as the search scope of the current search term; A search module, configured to search for recommended products within the search range based on the current search term; A scoring module, configured to calculate a scoring value for each of the recommended products based on the initial weights; The push module is used to sort the recommended products in descending order of their ratings and then push them to the user end.

10. The product search device according to claim 9, characterized in that The intention prediction module includes: A training corpus unit is used to obtain user information as training corpus, wherein the user information includes the material name of the product, the user's historical search terms and the associated information formed by browsing product information, and the user's quotation data information; A target data set unit is used to aggregate, remove duplicates and correct spelling errors on the training corpus in sequence to obtain a target data set; A training unit, configured to input a preset proportion of the target data set into a neural network model for training and output user intent; The intention recognition model unit is used to output the neural network model with the accuracy reaching the preset value as the intention recognition model when the accuracy of the neural network model reaches the preset value.

11. The product search device according to any one of claims 9 to 10, characterized in that: The scoring module includes: A scoring unit, configured to obtain at least two product rating items, and initial scores and item weights corresponding to each item, wherein the product rating items include part-of-speech items; an optimization unit, configured to use the initial weight as the item weight of the corresponding part-of-speech item of the corresponding recommended product; A calculation unit is used to calculate the rating value of the recommended product based on the initial score and item weight corresponding to the product rating item.

12. The product search device according to claim 11, characterized in that The product rating items of the rating unit further include product name items, brand items and attribute items, and the calculation unit further includes: The calculation subunit is used to obtain the rating value of the recommended product by adding the initial score of the product name item * the product weight + the initial score of the brand item * the brand weight + the initial score of the attribute item * the attribute weight + the initial score of the part-of-speech item corresponding to the initial weight * the part-of-speech weight, wherein the attribute weight < the product weight < the initial weight < the brand weight, and the attribute weight + the product weight + the brand weight = 1.

13. The product search device according to claim 9, characterized in that Also includes, Real-time detection module, used to obtain the user's browsing product information and browsing time in real time; The initial weight real-time updating module is used to update the initial weight of the current search term in real time based on the browsed product information and the browsing time.

14. The product search device according to claim 13, wherein The initial weight real-time update module includes: a weight reduction unit, configured to reduce the initial weights of a preset number of the current search terms ranked at the top to obtain real-time weights when the browsing time reaches a preset value and there is no browsing product information on the user terminal; A real-time updating unit is used to replace the initial weight of the corresponding current search term with the real-time weight.

15. The product search device according to claim 9, characterized in that Also includes, Statistics module, used to regularly collect statistics on the browsed products of each user terminal; The click-through rate module is used to calculate the click-through rate of the same product based on the product information browsed by each user terminal; The initial weight periodic updating module updates the initial weight of the current search term based on the click-through rate of the product.

16. The product search device according to claim 15, characterized in that The initial weight periodic update module includes: a weighting unit for increasing the initial weight of the current search term according to the click-through rate of the product to obtain a changed weight, wherein the greater the click-through rate of the product, the greater the initial weight of the current search term; The initial weight periodic updating unit is used to update the initial weight of the current search term using the changed weight.

17. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method according to any one of claims 1 to 8.

18. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images