E-commerce product search and recommendation method, device, electronic device and storage medium

Abstract

The present application relates to the field of Internet technology, and in particular to an e-commerce product search and recommendation method, device, electronic device and storage medium. The present application determines the search intent of the target user based on the search term input by the target user and the intent recognition model; searches for the target product label with the highest correlation with the search intent from the product label pool; searches for multiple recommended products that match the target product label from the product database, and recommends the multiple recommended products to the target user; the intent recognition model is trained based on the user behavior information of multiple sample users and the user label pool. In this way, the present application recommends products through deep learning of user behavior and user labels, which can improve the accuracy of user searches for products.

Description

Technical Field

[0001] The present application relates to the field of Internet technology, and in particular to an e-commerce product search and recommendation method, device, electronic device and storage medium. Background Art

[0002] With the rapid development of computer technology, more and more shopping apps are launched, and users can purchase goods conveniently through shopping apps, which improves users' shopping efficiency and shopping experience. Generally, shopping apps generally have a search function, that is, users can search for goods by entering search terms and other methods.

[0003] In traditional e-commerce platforms, search results are usually recommended based on search content, sales volume, price and other factors. However, in the current highly competitive market environment, users' patience is decreasing. If users cannot find the products they need quickly and accurately when searching, they may abandon the platform immediately. Therefore, how to provide accurate search results that meet user expectations is a technical problem that needs to be solved urgently.

[0004] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0005] In view of this, the embodiments of the present application at least provide an e-commerce product search recommendation method, device, electronic device and storage medium, which can improve the accuracy of user searches for products.

[0006] This application mainly includes the following aspects:

[0007] In a first aspect, an embodiment of the present application provides an e-commerce product search and recommendation method, the e-commerce product search and recommendation method comprising:

[0008] Get the search term entered by the target user;

[0009] Determine the search intent of the target user based on the search term and the intent recognition model; the intent recognition model is trained based on user behavior information and a user tag pool of multiple sample users;

[0010] Searching for a target product tag with the highest relevance to the search intent from a product tag pool; the relevance is determined based on a preset similarity algorithm;

[0011] A plurality of recommended commodities matching the target commodity label are searched from a commodity database, and the plurality of recommended commodities are recommended to the target user.

[0012] In a second aspect, the embodiment of the present application further provides an e-commerce product search and recommendation device, the e-commerce product search and recommendation device comprising:

[0013] An acquisition module is used to obtain the search term input by the target user;

[0014] A determination module, used to determine the search intent of the target user based on the search term and the intent recognition model; the intent recognition model is trained based on user behavior information of multiple sample users and a user tag pool;

[0015] A search module, used to search a target product tag with the highest relevance to the search intent from a product tag pool; the relevance is determined based on a preset similarity algorithm;

[0016] The recommendation module is used to search for multiple recommended commodities that match the target commodity label from the commodity database, and recommend the multiple recommended commodities to the target user.

[0017] In a third aspect, an embodiment of the present application further provides an electronic device, comprising: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and the machine-readable instructions are executed by the processor to execute the steps of the e-commerce product search and recommendation method described in the first aspect above.

[0018] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the e-commerce product search and recommendation method described in the first aspect or the first aspect are executed.

[0019] The e-commerce product search and recommendation method, device, electronic device and storage medium provided in the embodiment of the present application use the search term and intent recognition model based on the target user input to determine the target user's search intent; search the target product label with the highest correlation with the search intent from the product label pool; search for multiple recommended products that match the target product label from the product database, and recommend the multiple recommended products to the target user; the intent recognition model is trained based on the user behavior information of multiple sample users and the user label pool. In this way, the present application recommends products through deep learning of user behavior and user labels, which can improve the accuracy of user search for products.

[0020] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0022] Figure 1 A flowchart of an e-commerce product search and recommendation method provided in an embodiment of the present application is shown;

[0023] Figure 2 One of the functional module diagrams of an e-commerce product search and recommendation device provided in an embodiment of the present application is shown;

[0024] Figure 3 The second functional module diagram of an e-commerce product search and recommendation device provided in an embodiment of the present application is shown;

[0025] Figure 4 A schematic structural diagram of an electronic device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0026] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the drawings in the present application only serve the purpose of explanation and description and are not used to limit the scope of protection of the present application. In addition, it should be understood that the schematic drawings are not drawn in real proportion. The flowchart used in this application shows the operations implemented according to some embodiments of the present application. It should be understood that the operations of the flowchart can be implemented out of sequence, and the steps without logical context can be reversed in order or implemented simultaneously. In addition, those skilled in the art, under the guidance of the content of the present application, can add one or more other operations to the flowchart, or remove one or more operations from the flowchart.

[0027] In addition, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0028] In order to enable those skilled in the art to use the contents of this application, the following implementation methods are given in combination with the specific application scenario of "product search recommendation". For those skilled in the art, the general principles defined here can be applied to other embodiments and application scenarios without departing from the spirit and scope of this application.

[0029] The following methods, devices, electronic devices or computer-readable storage media of the embodiments of the present application can be applied to any scenario requiring product search and recommendation. The embodiments of the present application are not limited to specific application scenarios. Any scheme using the XX method and device provided in the embodiments of the present application is within the protection scope of the present application.

[0030] It is worth noting that with the rapid development of computer technology, more and more shopping apps are launched, and users can purchase goods conveniently through shopping apps, which improves users' shopping efficiency and shopping experience. Generally, shopping apps generally have a search function, that is, users can search for goods by entering search terms and other methods.

[0031] Before this application was filed, in traditional e-commerce platforms, search results were usually recommended based on factors such as search content, sales volume and price. Traditional search methods include keyword search, category browsing, price screening, brand screening, sales volume ranking, evaluation ranking, comprehensive ranking, attribute ranking, etc. However, in the current highly competitive market environment, users' patience is decreasing. If users cannot find the required products quickly and accurately when searching, they may abandon the platform immediately. Therefore, how to provide accurate search results that meet user expectations is a technical problem that needs to be solved urgently.

[0032] In response to the above problems, the embodiment of the present application determines the search intent of the target user based on the search term input by the target user and the intention recognition model; searches for the target product label with the highest correlation with the search intent from the product label pool; searches for multiple recommended products that match the target product label from the product database, and recommends the multiple recommended products to the target user; the intention recognition model is trained based on the user behavior information of multiple sample users and the user label pool. In this way, the present application recommends products through deep learning of user behavior and user labels, which can improve the accuracy of user searches for products.

[0033] To facilitate the understanding of the present application, the technical solution provided by the present application is described in detail below in conjunction with specific embodiments.

[0034] Figure 1 This is a flow chart of an e-commerce product search and recommendation method provided in an embodiment of the present application. Figure 1 As shown, the e-commerce product search and recommendation method provided in the embodiment of the present application includes the following steps:

[0035] S101: Acquire a search term input by a target user.

[0036] In a specific implementation, the e-commerce product search and recommendation method provided in the embodiment of the present application can be applied to an e-commerce platform. Each user can input search terms related to the product to be searched through the e-commerce platform, and the e-commerce platform can recommend products to the corresponding user based on the search terms.

[0037] S102: Determine the search intent of the target user based on the search term and the intent recognition model.

[0038] The intent recognition model is trained based on user behavior information and a user tag pool of multiple sample users.

[0039] In the specific implementation, the search intent of the target user is determined through the search terms input by the target user and the trained intent recognition model. Specifically, the intent recognition model is trained through the user behavior information and user tag pool of a large number of sample users. Each sample user corresponds to its own user tag pool, which is composed of multiple user tags. The user tag pool of each sample user can be determined by the historical behavior information of the sample user. In this way, when the target user searches for a certain product, the user behavior information of the target user can be analyzed through the intent recognition model, and analyzed based on the user tags accumulated by the target user himself, so as to make targeted recommendations to the target user based on the search terms.

[0040] Here, the intent recognition model can be a deep learning model, such as a convolutional neural network model (CNN), a recurrent neural network model (RNN), a network structure model based on a self-attention mechanism (Transformer), etc.

[0041] S103: Searching for a target product tag with the highest relevance to the search intent from a product tag pool.

[0042] The relevance is determined based on a preset similarity algorithm.

[0043] In a specific implementation, the search intent of the target user can be matched with the product tag pool, and a preset similarity algorithm can be used to find the most relevant target product tag M1=TagMatching(I, S), where I is used to characterize the search intent and S is used to characterize the product tag pool. The embodiment of the present application can dynamically match the product tags of the products to be searched by the target user through the real-time identification of the search intent of the target user, so that the product recommendations can be made more accurately.

[0044] The preset similarity algorithm may be cosine similarity or other similarity algorithms.

[0045] S104: Searching for a plurality of recommended commodities that match the target commodity label from a commodity database, and recommending the plurality of recommended commodities to the target user.

[0046] In a specific implementation, after determining the target product label of the product to be searched by the target user, multiple recommended products that match the target product label can be searched from the product database. Here, the product database includes product data of each product of various product types on the e-commerce platform. In this way, the searched multiple recommended products can be recommended to the target user.

[0047] Among them, the embodiment of the present application constructs a closed-loop recommendation as a whole, from user behavior analysis to product recommendation, and then to iterative update of tags.

[0048] In the embodiment of the present application, the search intent of the target user is determined based on the search term input by the target user and the intention recognition model; the target product label with the highest correlation with the search intent is searched from the product label pool; multiple recommended products matching the target product label are searched from the product database, and the multiple recommended products are recommended to the target user; the intention recognition model is trained based on the user behavior information of multiple sample users and the user label pool. In this way, the present application recommends products through deep learning of user behavior and user labels, which can improve the accuracy of user search for products.

[0049] In a possible implementation, determining the search intent of the target user based on the search term and the intent recognition model in S102 includes the following steps:

[0050] Step 1021: preprocess the search words, and convert the preprocessed search words into word vectors;

[0051] Step 1022: Input the word vector into a sequence model to extract keywords, and obtain keywords corresponding to the word vector;

[0052] Step 1023: Input the keyword into the intent recognition model to perform intent recognition to obtain the search intent of the target user.

[0053] In a specific implementation, after obtaining the search terms input by the target user, the search terms can be preprocessed first, such as cleaning and standardization, including removing stop words, stem extraction, conversion to lowercase, etc., and then the preprocessed search terms are converted into word vectors E(U); then, the word vectors are input into a sequence model to extract keywords K=KeywordExtraction(E(U)); here, the sequence model is such as a deep learning model (BidirectionalLongShort-TermMemory, Bi-LSTM), a network structure of a self-attention mechanism (Transformer); further, the extracted keywords are passed through an intent recognition model to understand the target user's search intent I=IntentRecognition(K), where K represents keywords and I represents search intent. In this way, the embodiment of the present application can use natural language processing (Natural Language Processing, NLP) and machine learning to understand the user's search intent and can match user preferences with keywords.

[0054] In a possible implementation manner, the user tag in the user tag pool of the target user is determined according to at least one of the following methods:

[0055] Acquire the registration information of the target user, and determine the average unit price of user consumption in the area associated with the IP address included in the registration information as a user tag;

[0056] Recording the historical product click behavior of the target user, and determining the price stage of the candidate product targeted by the historical product click behavior among similar products as a user tag;

[0057] Record the target user's historical product click behavior on the search results, and determine the product type, product price range, and product brand of the candidate product targeted by the historical product click behavior as a user tag;

[0058] Recording the historical product browsing behavior of the target user, and determining the product attributes of the candidate products targeted by the historical product browsing behavior as user tags;

[0059] Recording the historical search behavior of the target user, and performing natural language processing on the search content corresponding to the historical search behavior to obtain a content tag, and determining it as a user tag;

[0060] The search content of the target user is received. If it is identified that the search content contains typos, the search content is corrected for synonyms using a deep learning algorithm and a product tag pool, and the search content obtained after the correction is determined as a user tag.

[0061] In a specific implementation, the embodiments of the present application can generate user tags based on the user's historical behavior. The user's historical behavior may include historical registration behavior, historical search behavior, historical commodity click behavior, historical commodity browsing behavior, etc. Specifically, the user tags in the user tag pool generated by the embodiments of the present application are mainly divided into two stages. The first stage can be the establishment of user behavior tendency, and the second stage can be the establishment of user search optimization. Specifically, for the first stage, different user tags can be generated corresponding to the historical registration behavior and historical commodity click behavior of the target user. For example, according to the IP address included in the registration information of the target user, the average unit price of user consumption in this area is obtained as one of the references for the weight of the recommended commodity amount, that is, the average unit price of user consumption is determined as a user tag; through the historical commodity click behavior of the target user on the candidate commodities, the historical commodity click behavior is recorded, and the price stage of the candidate commodity clicked by the target user among the same type of commodities is used as another user tag; for the second stage, different user tags can be generated corresponding to the historical commodity browsing behavior, historical search behavior, and historical commodity click behavior for the search results of the target user. For example, when the target user enters a search content, the content tag obtained by performing natural language processing on the search content entered by the target user is determined as a user tag. For example, a user who enters "red pullover" will be willing to look at sweaters and pullovers. Tags such as "red" and "pullover" after natural language processing are added to the user tag pool. In this way, even if the target user does not include the word "pullover" in the search content later, "pullover styles" are preferably recommended for clothes such as sweaters. Similarly, it also applies to colors and the ability to understand the changes of red; in addition, sometimes there may be typos in the search content of the target user. At the same time, with the help of deep learning and combined with the commodity tag pool, the content entered by the target user is corrected for synonyms. For example, if the target user searches for "sleep already" or "sleep dependently", then through deep learning, it can be understood that the user wants to search for "pajamas". If the user enters "sleepiness", then deep learning may understand it as a sleep aid product, and the corrected search content is also determined as a user tag; record the historical commodity click behavior of the user clicking on the search results. For example, when the user searches for "water", the existing commodities on the platform will be listed: soda water, mineral water, purified water, and the first screen of the search results is different types of water. First, it is judged what kind of water the user wants. When the user clicks on soda water, the price range tag, brand positioning tag, etc. of this commodity are added to the user tag pool at the same time. In addition to the tags, the tag quantity is accumulated when recording the tags again, so as to optimize the data analysis results more.

[0062] Here, the embodiment of the present application allows typos and synonyms in the search terms entered by the user, that is, it can accommodate the user's search input content, and the search hit rate can be improved by using natural language processing. In addition, by making recommendations based on these user tags and search terms, the personalization of the recommendations can be improved. In addition, user tags can be automatically updated according to changes in user behavior to maintain the accuracy of real-time recommendations.

[0063] In a possible implementation manner, the product tags in the product tag pool are determined according to at least one of the following methods:

[0064] If any of the sample users does not click on any search result after searching for a product, the candidate products clicked by the sample user in the second search in the search session are recorded, and the search content of the second search by the sample user is determined as a product label;

[0065] For any of the sample user's search content, if the sample user clicks on a candidate product obtained through the search content, adding the search content as a product tag to the candidate product;

[0066] For any candidate product obtained by clicking on the search content by the sample user, the search content is split according to vocabulary to obtain at least two split words, and each split word is added as a product label to other products belonging to the same category as the candidate product.

[0067] In a specific implementation, a product label can be added to each product on the e-commerce platform. Specifically, if a user searches for a product but does not click on any search result, the user will conduct a secondary search in this search session. When a candidate product is clicked on any search result, the user's search content will be added to the product label pool as a product label for use by other users when searching. For the expansion of the product label pool, if a user clicks on a candidate product obtained through the search content, the search content will be added to the candidate product as a product label. Furthermore, the search content can be split according to vocabulary to obtain at least two split words, and each split word will be added as a product label to other products that belong to the same category as the candidate product. In this way, not only a single product but also the attached labels will be split, and the keywords will be matched with the product label pool of similar products. If they are similar, the label will be extended to products in the same category. For example, the user searches for "happy water" and finds candidate products such as "juice", "carbonated drinks", and "tea drinks". The reason is that when the user first searches for "happy water", only the result of "water" will be matched to the user. "Carbonated drinks" will be recommended through substitute products. After the user clicks, the "happy water" label will be applied to "carbonated drinks". At the same time, "juice", "tea drinks" and "carbonated drinks" are in the same category, so "happy water" will be split into "happy" and "water" according to the vocabulary. Again, check whether the word "water" is included in the label pools of "juice" and "tea drinks", and there is no "happy". Then the "happy" label will be applied to "juice" and "tea drinks". In this way, the product label expansion is completed, and the label expansion of similar products and substitute products is also completed.

[0068] Here, the embodiment of the present application not only adds user tags for each user in the e-commerce platform, but also adds product tags for each product in the e-commerce platform, and can automatically improve product tags to improve the coverage and accuracy of tags. In this way, through the analysis of big data based on the whole platform, that is, through the two-way dynamic learning of tags based on user behavior, the search results can be expanded by using self-learning two-way tag matching, which can improve the transaction rate.

[0069] In a possible implementation, the step of recommending the plurality of commodities to be recommended to the target user in S104 includes the following steps:

[0070] Step 1041, for any of the to-be-recommended products, determine the factor score and factor weight of the to-be-recommended product under each ranking factor; each ranking factor is determined based on relevance and popularity;

[0071] Step 1042, determining the product score of the product to be recommended for the target user according to the factor score of each ranking factor of the product to be recommended and the factor weight of each ranking factor; the ranking factor is any one of the following factors: platform activity factor, coupon value factor, user preference factor, product click-through rate factor, product transaction volume factor, user historical purchase price range factor, user social situation factor;

[0072] Step 1043, ranking the multiple products to be recommended according to the product score of each product to be recommended, and displaying the multiple products to be recommended to the target user in the form of a product list according to the ranking result.

[0073] In a specific implementation, the embodiment of the present application will sort the search results (multiple products to be recommended). Specifically, the display principle is based on the user search, matching the content of the products, and then intelligently sorting the multiple products to be recommended based on platform activity factors, coupon value factors, user preference factors, product click-through rate factors, product transaction volume factors, user historical purchase price range factors, and user social situation factors. Finally, based on in-depth analysis of user behavior and search similarities, the products to be recommended are interspersed in between.

[0074] Here, we can define the weight vector ,in is the number of ranking factors considered, Is the sorting factor The factor weights, The value is greater than or equal to and less than or equal to A positive integer. For any product to be recommended Product score , you can use the recommended product Under each sorting factor The factor score and the product of the factor weight of each ranking factor are summed up to determine, specifically, any product to be recommended For target users The intelligent ranking score, that is, the score of the recommended product can be expressed as: ,in, Is a function used to calculate any recommended product In any sorting factor The factor scores under this ranking factor With target users Here, the ranking factor It may be factors such as platform activity, coupon value, user preference, product click-through rate, product transaction volume, user's historical purchase price range, and user social status.

[0075] Specifically for each ranking factor, the factor score under each ranking factor is defined as follows:

[0076] :Products to be recommended The factor score under the platform activity factor may be related to the engagement of the target user u;

[0077] :Products to be recommended factor scores under the coupon value factor;

[0078] :Products to be recommended Factor scores under the user preference factor;

[0079] :Products to be recommended Factor score under the product click-through rate factor;

[0080] :Products to be recommended Factor score under the commodity volume factor;

[0081] :Products to be recommended Factor score under the user's historical purchase price range factor;

[0082] :Products to be recommended The factor score under the user's social situation factor.

[0083] In a possible implementation, if the ranking factor is a platform activity factor, the factor score of the to-be-recommended product under the platform activity factor is determined according to the following formula: ;

[0084] in, Recommended products the factor score under the said platform activity factor, is the benchmark score for the platform activity factor, Adjust the index for the activity score, Score platform activities, For the target user For the recommended product The participation score of the platform activity score is calculated as follows: , Base score for the activity, is the activity intensity coefficient, which indicates the direct impact of the activity on the score. is the total number of days the activity lasts, is the current time point, starting from the start of the activity. Is greater than or equal to 1 and less than or equal to A positive integer, is the first time decay coefficient, which is used to adjust the decay of activity score over time. is the base of the natural logarithm; the calculation formula for the participation score is , For the target user The recommended products The total number of interactions between For the The weight of the interaction. These weights can vary depending on the type of interaction, that is, different weights can be assigned according to the type of interaction. For the The basic score of each interaction, is the total duration of the time window in which the user interacted (for example, the last 30 days), For the The time at which the interaction occurred. is the second time decay coefficient, which is used to adjust the decay of the user interaction score over time. That is, the decay coefficient can be used to indicate that the impact of target user interactions on the engagement score decreases over time.

[0085] Among them, the platform activity score The calculation formula takes into account the decay effect of activities over time, that is, the contribution to the score is the largest at the beginning of the activity, and the contribution gradually decreases over time. For example, if an activity lasts for 7 days, the recommended products Score of daily platform activity ;

[0086] The calculation formula for the engagement score is The importance of different types of interactions is taken into account (via weights ) and the timeliness of user interaction (via the second time decay coefficient ), weight Different scores can be assigned to different interaction types based on business logic. For example, sharing a product may receive a higher weight than clicking on a product. Products to be recommended in the past 30 days There were 5 interactions, including 3 clicks ( )、1 collection( ) and 1 share ( ), which can be calculated as , this formula weights the scores of all interactions and takes time decay into account to get the target user About Products Total engagement score.

[0087] In a possible implementation, if the ranking factor is the coupon value factor, the recommended product is determined according to the following formula Factor scores under the stated coupon value factor:

[0088] ;

[0089] in, The recommended product the factor score under said coupon value factor, is the benchmark score for the coupon value factor, Adjust the index for the coupon score, is the coupon value score; the calculation formula of the coupon value score is: , For the target user, is the product to be recommended, For the target user The total number of coupons available, For the target user Available The face value of the coupons, is a weight function used to determine the coupon value Calculate the target user The influence of purchase intention, this function takes into account factors such as the denomination of the coupon and the usage threshold.

[0090] Here, use To indicate that the coupon may be treated as a recommended product has a negative impact on attractiveness.

[0091] Here, the weight function Can have different forms:

[0092] ;

[0093] in, is a threshold, indicating that only when the coupon value is greater than this threshold will it have a significant impact on the user's purchase intention; is a coefficient less than 1 that adjusts the effect of smaller denomination coupons. For example, if a user has 3 available coupons, denominations of 10 yuan, 5 yuan, and 20 yuan, and Yuan, , which can be calculated as , this formula adds up the weighted denominations of all coupons to get the target user For products to be recommended The coupon value score.

[0094] In a possible implementation, if the ranking factor is a user preference factor, the factor score of the to-be-recommended product under the user preference factor is determined according to the following formula:

[0095] ;

[0096] in, Recommended products factor scores under the user preference factors, is the benchmark score of the user preference factor, Adjust the index for the preference score, is the user preference score; the calculation formula of the user preference score is: , For users to be recommended, For products to be recommended, For target users The number of favorite tags, Recommended products The number of attributes, For target users No. Favorite tags, Recommended products No. properties, is a matching function used to quantize labels and properties The degree of match between them.

[0097] Here, the matching function Can have different forms:

[0098] ; This function can return three levels of match: complete match, partial match, and no match. For example, if the target user There are three favorite tags , , , products to be recommended There are two properties and , it can be calculated as follows: ; This formula compares the matching degree of each preference tag of the target user with all the attributes of the recommended product, takes the maximum value, and then sums up all the tags to obtain the user preference score of the target user for the recommended product.

[0099] In a possible implementation, if the ranking factor is a product click-through rate factor, the factor score of the product to be recommended under the product click-through rate factor is determined according to the following formula:

[0100] ;

[0101] in, The recommended product The factor score under the product click-through rate factor, is the benchmark score of the product's click-through rate factor, Adjust the index for the CTR score, Recommended products Click-through rate, is the average click-through rate of all products on the platform; the recommended products The formula for calculating the click rate is , For the recommended product X in Number of clicks received for impressions, For the The timestamp of the impression, is the end time of the time window under consideration (for example, the last 30 days), is the total number of product displays within the time window, is the decay rate, which is used to adjust the impact of old click events on the click rate. It can also quantify the impact of time on the click rate. is the base of the natural logarithm; the calculation formula for the average click-through rate of all platform products is , N4 is the total number of commodities considered in the time window, For the The total number of clicks on products in the time window, For the The total number of product impressions.

[0102] Here are the recommended products The formula for calculating the click-through rate By using exponential decay, we give higher weight to newer click events while ensuring that the calculation of click rate is normalized based on the number of impressions. 100 impressions in the last 30 days, with the following clicks: , and the timestamps of each display are , the calculation formula is ; Calculation formula for average click-through rate of all platform products The ratio between the sum of clicks and the sum of impressions of all products is calculated to get an average click-through rate, which can be used as a benchmark for evaluating the relative attractiveness of products. For example, if there are 100 products on an e-commerce platform in the last 30 days, their clicks and impressions are , the calculation method is: .

[0103] In a possible implementation, if the ranking factor is a commodity transaction volume factor, the factor score of the commodity to be recommended under the commodity transaction volume factor is determined according to the following formula:

[0104] ;

[0105] in, The recommended product The factor score under the commodity volume factor, is the benchmark score of the commodity volume factor, Adjust the index for volume score, Recommended products The transaction volume, is the average transaction volume of all products on the platform; the recommended products The calculation formula for the trading volume is , Recommended products In time The transaction volume, is the size of the time window to consider (e.g., the last 30 days or a quarter), is the total number of time units (such as days or hours) within the time window; the calculation formula for the average transaction volume of all platform products is , For the Products at time The transaction volume, is the size of the time window to consider (e.g., the last 30 days), is the total number of time units (such as days or hours) within the time window, is the total number of items considered in the time window.

[0106] Here are the recommended products The calculation formula of trading volume Add up the transaction volume of each time unit in the time window to get the recommended product The total transaction volume within the time window, for example, if you want to calculate the recommended products The total trading volume in the last 30 days, and the daily trading volume in these 30 days is , then calculate ; The calculation formula for the average transaction volume of all platform products , first calculate the sum of the transaction volume of each product in all time units in the time window, and then find the average of all these sums to get the average transaction volume. For example, if there are 100 products on an e-commerce platform in the last 30 days, and the daily transaction volumes of these products are arrive , calculated as:

[0107] .

[0108] In a possible implementation, if the ranking factor is the user's historical purchase price range factor, the factor score of the recommended product under the user's historical purchase price range factor is determined according to the following formula:

[0109] ;

[0110] in, The recommended product The factor score under the user's historical purchase price range factor, is the benchmark score of the user's historical purchase price range factor, Adjust the index for the price range score, For target users Historical purchase price range and recommended products The price matching score; the matching score is calculated as follows: , For target users The first The price of the item purchased. For target users The total number of historical purchase records, Recommended products The current price of is a price matching function that quantifies the degree of match between two prices.

[0111] Here, by defining a price matching function To quantify the target user's historical purchase price With the current price of the product The degree of match between them.

[0112] Price Matching Function There are different forms, such as , is a preset threshold value, indicating that the price difference is within an acceptable range; Indicates the absolute difference between the user's historical purchase price and the current price of the product. For example, if the target user There are 5 historical purchase records, the prices are , Recommended products The current price is , .

[0113] Among them, this formula performs a weighted sum of the matching degree of all historical purchase prices of the user with the current price of the product, and then takes the average to obtain the target user's fitness score for the product price.

[0114] In a possible implementation, if the ranking factor is the user's social situation factor, the factor score of the to-be-recommended product under the user's social situation factor is determined according to the following formula:

[0115] ;

[0116] in, The recommended product the factor score under the user's social situation factor, is the benchmark score of the user's social situation factor, Adjust the index for user social interaction. For target users Social network treatment of recommended products Impact score; target users Social network treatment of recommended products The formula for the impact score is , For users Treat recommended products Like score, based on user Treat recommended products It is calculated based on the clicks, purchases, shares or comments of For target users Total number of friends in social network, For users For target users The weight of influence in a social network can be determined based on factors such as the strength of their relationships and the frequency of interaction.

[0117] For example, if the user Treat recommended products The preference score of a person is based on the number of times they purchased the product. It can be the number of purchases. If the target user Have 5 social friends who recommend products The preference scores are , and the corresponding weights are The calculation is:

[0118] ; This formula treats each friend in the user's social network as a recommended product The preference scores of the products are multiplied by their weights, and then these products are added up to get the recommended products. Treating recommended users Social influence score.

[0119] Based on the same application concept, the embodiment of the present application also provides an e-commerce product search and recommendation device corresponding to the e-commerce product search and recommendation method provided in the above embodiment. Since the principle of solving the problem by the device in the embodiment of the present application is similar to the e-commerce product search and recommendation method in the above embodiment of the present application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0120] like Figure 2 , 3 As shown, Figure 2 This is one of the functional module diagrams of an e-commerce product search and recommendation device 200 provided in an embodiment of the present application. Figure 3 This is a second functional module diagram of an e-commerce product search and recommendation device 200 provided in an embodiment of the present application. Figure 2 As shown, the e-commerce product search and recommendation device 200 includes:

[0121] The acquisition module 210 is used to acquire the search term input by the target user;

[0122] A determination module 220 is used to determine the search intent of the target user based on the search term and the intent recognition model; the intent recognition model is trained based on user behavior information of multiple sample users and a user tag pool;

[0123] A search module 230 is used to search a target product tag with the highest relevance to the search intent from a product tag pool; the relevance is determined based on a preset similarity algorithm;

[0124] The recommendation module 240 is used to search for a plurality of recommended commodities that match the target commodity tag from the commodity database, and recommend the plurality of recommended commodities to the target user.

[0125] In a possible implementation, Figure 3 As shown, the determination module 220 includes an intention determination unit 221; the intention determination unit 221 is specifically used to determine the search intention of the target user according to the following steps:

[0126] Preprocessing the search words, and converting the preprocessed search words into word vectors;

[0127] Input the word vector into a sequence model to extract keywords, and obtain keywords corresponding to the word vector;

[0128] The keyword is input into the intent recognition model for intent recognition to obtain the search intent of the target user.

[0129] In a possible implementation, Figure 3 As shown, the recommendation module 240 is specifically used to recommend the multiple recommended commodities to the target user according to the following steps:

[0130] For any of the products to be recommended, determine the factor score and factor weight of the product to be recommended under each ranking factor; each ranking factor is determined based on relevance and popularity;

[0131] Determine the product score of the product to be recommended for the target user according to the factor score of each ranking factor of the product to be recommended and the factor weight of each ranking factor; the ranking factor is any one of the following factors: platform activity factor, coupon value factor, user preference factor, product click-through rate factor, product transaction volume factor, user historical purchase price range factor, user social situation factor;

[0132] The multiple commodities to be recommended are ranked according to the commodity scores of the commodities to be recommended, and the multiple commodities to be recommended are displayed to the target user in the form of a commodity list according to the ranking result.

[0133] Based on the same application concept, see Figure 4 As shown, it is a structural diagram of an electronic device 400 provided in an embodiment of the present application, including: a processor 410, a memory 420 and a bus 430, the memory 420 stores machine-readable instructions executable by the processor 410, when the electronic device 400 is running, the processor 410 and the memory 420 communicate through the bus 430, and the machine-readable instructions are executed by the processor 410 when running to perform the steps of the e-commerce product search and recommendation method as described in any of the above embodiments.

[0134] Based on the same application concept, an embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the e-commerce product search and recommendation method provided in the above embodiment are executed.

[0135] Specifically, the storage medium can be a general storage medium, such as a mobile disk, a hard disk, etc. When the computer program on the storage medium is run, it can execute the above-mentioned e-commerce product search recommendation method, and make product recommendations through deep learning of user behavior and user tags, which can improve the accuracy of user search for products.

[0136] In the embodiment of the present application, the computer program can also execute other machine-readable instructions when run by the processor to execute other methods described in the embodiment. For the specific execution method steps and principles, please refer to the description of the embodiment, which will not be repeated here.

[0137] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0138] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0139] In addition, each functional unit in the embodiments provided in the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0140] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program codes.

[0141] It should be noted that similar numbers and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are only used to distinguish the description and are not to be understood as indicating or implying relative importance.

[0142] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the above-mentioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-mentioned embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application. They should all be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. An e-commerce product search and recommendation method, characterized in that: The e-commerce product search and recommendation method comprises: Get the search term entered by the target user; Determine the search intent of the target user based on the search term and the intent recognition model; the intent recognition model is trained based on user behavior information and a user tag pool of multiple sample users; Searching for a target product tag with the highest relevance to the search intent from a product tag pool; the relevance is determined based on a preset similarity algorithm; Searching for multiple recommended commodities that match the target commodity label from a commodity database, and recommending the multiple recommended commodities to the target user; The recommending the plurality of commodities to be recommended to the target user comprises: For any of the products to be recommended, determine the factor score and factor weight of the product to be recommended under each ranking factor; each ranking factor is determined based on relevance and popularity; Determine the product score of the product to be recommended for the target user according to the factor score of each ranking factor of the product to be recommended and the factor weight of each ranking factor; the ranking factor is a platform activity factor; According to the commodity scores of the commodities to be recommended, the plurality of commodities to be recommended are ranked for recommendation, and the plurality of commodities to be recommended are displayed to the target user in the form of a commodity list according to the ranking results; If the ranking factor is a platform activity factor, the factor score of the recommended product under the platform activity factor is determined according to the following formula: ； in, Products to be recommended the factor score under the said platform activity factor, is the benchmark score for the platform activity factor, Adjust the index for the activity score, Score platform activities, For the target user For the recommended product The participation score of the platform activity score is calculated as follows: , Base score for the activity, is the activity intensity coefficient, is the total number of days the activity lasts, is the current time point, Is greater than or equal to 1 and less than or equal to A positive integer, is the first time attenuation coefficient, is the base of the natural logarithm; the calculation formula for the participation score is , For the target user The recommended products The total number of interactions between For the The weight of the interaction, For the The basic score of each interaction, is the total duration of the time window in which the user interacts, For the The time at which the interaction occurred. is the second time attenuation coefficient.

2. The e-commerce product search and recommendation method according to claim 1, characterized in that: The determining the search intent of the target user based on the search term and the intent recognition model includes: Preprocessing the search words, and converting the preprocessed search words into word vectors; Input the word vector into a sequence model to extract keywords, and obtain keywords corresponding to the word vector; The keyword is input into the intent recognition model for intent recognition to obtain the search intent of the target user.

3. The e-commerce product search and recommendation method according to claim 1, characterized in that: Determine the user tag in the user tag pool of the target user in at least one of the following ways: Acquire the registration information of the target user, and determine the average unit price of user consumption in the area associated with the IP address included in the registration information as a user tag; Recording the historical product click behavior of the target user, and determining the price stage of the candidate product targeted by the historical product click behavior among similar products as a user tag; Record the target user's historical product click behavior on the search results, and determine the product type, product price range, and product brand of the candidate product targeted by the historical product click behavior as a user tag; Recording the historical product browsing behavior of the target user, and determining the product attributes of the candidate products targeted by the historical product browsing behavior as user tags; Recording the historical search behavior of the target user, and performing natural language processing on the search content corresponding to the historical search behavior to obtain a content tag, and determining it as a user tag; The search content of the target user is received. If it is identified that the search content contains typos, the search content is corrected for synonyms using a deep learning algorithm and a product tag pool, and the search content obtained after the correction is determined as a user tag.

4. The e-commerce product search and recommendation method according to claim 1, characterized in that: Determine the product tags in the product tag pool according to at least one of the following methods: If any of the sample users does not click on any search result after searching for a product, the candidate products clicked by the sample user in the second search in the search session are recorded, and the search content of the second search by the sample user is determined as a product label; For any of the sample user's search content, if the sample user clicks on a candidate product obtained through the search content, adding the search content as a product tag to the candidate product; For any candidate product obtained by clicking on the search content by the sample user, the search content is split according to vocabulary to obtain at least two split words, and each split word is added as a product label to other products belonging to the same category as the candidate product.

5. The e-commerce product search and recommendation method according to claim 1, characterized in that: If the ranking factor is a coupon value factor, the factor score of the recommended product under the coupon value factor is determined according to the following formula: ； in, The recommended product the factor score under said coupon value factor, is the benchmark score for the coupon value factor, Adjust the index for the coupon score, is the coupon value score; the calculation formula of the coupon value score is: , For the target user, is the product to be recommended, For the target user The total number of coupons available, For the target user Available The face value of the coupons, is a weight function used to determine the coupon value Calculate the target user The impact of purchasing intention.

6. An e-commerce product search and recommendation device, characterized in that: The e-commerce product search and recommendation device comprises: An acquisition module is used to obtain the search term input by the target user; A determination module, used to determine the search intent of the target user based on the search term and the intent recognition model; the intent recognition model is trained based on user behavior information of multiple sample users and a user tag pool; A search module, used to search a target product tag with the highest relevance to the search intent from a product tag pool; the relevance is determined based on a preset similarity algorithm; A recommendation module, used to search for a plurality of recommended commodities matching the target commodity label from a commodity database, and recommend the plurality of recommended commodities to the target user; The recommending the plurality of commodities to be recommended to the target user comprises: For any of the products to be recommended, determine the factor score and factor weight of the product to be recommended under each ranking factor; each ranking factor is determined based on relevance and popularity; Determine the product score of the product to be recommended for the target user according to the factor score of each ranking factor of the product to be recommended and the factor weight of each ranking factor; the ranking factor is a platform activity factor; According to the commodity scores of the commodities to be recommended, the plurality of commodities to be recommended are ranked for recommendation, and the plurality of commodities to be recommended are displayed to the target user in the form of a commodity list according to the ranking results; If the ranking factor is a platform activity factor, the factor score of the recommended product under the platform activity factor is determined according to the following formula: ； in, Products to be recommended the factor score under the said platform activity factor, is the benchmark score for the platform activity factor, Adjust the index for the activity score, Score platform activities, For the target user For the recommended product The participation score of the platform activity score is calculated as follows: , Base score for the activity, is the activity intensity coefficient, is the total number of days the activity lasts, is the current time point, Is greater than or equal to 1 and less than or equal to A positive integer, is the first time attenuation coefficient, is the base of the natural logarithm; the calculation formula for the participation score is , For the target user The recommended products The total number of interactions between For the The weight of the interaction, For the The basic score of each interaction, is the total duration of the time window in which the user interacts, For the The time at which the interaction occurred. is the second time attenuation coefficient.

7. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the memory communicate through the bus. When the machine-readable instructions are run by the processor, the steps of the e-commerce product search and recommendation method as described in any one of claims 1 to 5 are executed.

8. 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 the processor, the steps of the e-commerce product search and recommendation method according to any one of claims 1 to 5 are executed.

Images