Search trending words determination method, apparatus, and electronic device

Abstract

Disclosed in the present application are a search trending words determination method, an apparatus, and an electronic device. The present application relates to the field of intelligent trending words recommendation, and the method comprises: obtaining object behavior data of a target account within a preset time period (S102); performing merging on object information and the object behavior data according to an object identifier, and obtaining object details (S104); determining the account type of the target account and a trending words recall model corresponding to the account type on the basis of the number of times an object behavior was performed by the target account within a preset time period (S106); using the trending words recall model to obtain an object recommendation result for the target account on the basis of the object details (S108); and obtaining search trending words for the target account on the basis of the object recommendation result (S110).

Description

Search hot word determination method, device and electronic device

[0001] Related applications

[0002] This application claims priority to Chinese patent application number 2023114260319, filed on October 30, 2023, entitled “Method, device and electronic device for determining hot search words,” the entire text of which is hereby incorporated by reference. Technical Field

[0003] The present application relates to the field of intelligent hot word recommendation, and more specifically, to a method, device and electronic device for determining search hot words. Background Art

[0004] Currently, users are paying more and more attention to information-based interactive experience. Business activities are usually accompanied by a large number of user requests for business service access, which puts significant pressure on service resources. In order to alleviate the impact of the surge in user access on services, it is usually necessary to prepare more service resources, which will result in higher cost expenditures.

[0005] In e-commerce scenarios, traditional search keyword recommendation methods often rely on manual configuration or calculation of product interest weights based on historical user behavior to obtain recommended vocabulary information. This information is then used to provide product recommendations to accounts on e-commerce search pages. However, these traditional e-commerce keyword recommendation methods often rely on product data annotation, which requires significant human effort. Furthermore, these traditional recommendation methods typically use a single keyword recall model to predict search keyword recommendation results for all accounts, resulting in low prediction accuracy and poor applicability. Currently, no effective solutions have been proposed to address these issues. Summary of the Invention

[0006] Embodiments of the present application provide a method, device, and electronic device for determining a hot search term.

[0007] According to one aspect of an embodiment of the present application, a method for determining search hot words is provided, including: obtaining object behavior data of a target account within a preset time period, wherein the object behavior data includes: the number of object behaviors of the target account within the preset time period, the time when the object behavior occurs and the trigger type of the object behavior, and the preset time period is a period of a predetermined length before the current time; merging the object information and the object behavior data according to the object identifier to obtain object details, wherein the object details are used to indicate the correspondence between the object behavior of the target account and the object information; determining the account type of the target account and a hot word recall model corresponding to the account type based on the number of object behaviors of the target account within the preset time period; obtaining an object recommendation result of the target account based on the object details and the hot word recall model; and obtaining a search hot word of the target account based on the object recommendation result.

[0008] Optionally, the account type of the target account and the hot word recall model corresponding to the account type are determined based on the number of times the target account performs the object behavior within the preset time period, including: when the number of triggering times is greater than a first preset number, the account type is determined to be a first type of account, and the corresponding hot word recall model is a time series recall model; when the number of triggering times is less than or equal to the first preset number and greater than a second preset number, the account type is determined to be a second type of account, and the corresponding hot word recall model is a Newton's cooling law model.

[0009] Optionally, the method also includes: when the number of triggering times is less than or equal to the second preset number, obtaining the search record hot words corresponding to the accounts included in the e-commerce platform, wherein the search record hot words are determined based on the historical search records of the accounts included in the e-commerce platform; based on the search record hot words corresponding to the accounts included in the e-commerce platform, determining the search hot words of the target account.

[0010] Optionally, in the case where the account type is the first type of account, the object recommendation result of the target account is obtained based on the object details and the hot word recall model is adopted, including: based on the object details, obtaining the model input features corresponding to each moment when the object behavior of the target account occurs within the preset time period, wherein the model input features include: feature triples, and at least one of the following: object category, object price and sales volume, object marketing score, object title, target account information, the feature triples include target account identifier, object identifier and timestamp information; based on the model input features corresponding to each moment when the object behavior of the target account occurs, the temporal recall model is adopted to obtain the object recommendation result of the target account.

[0011] Optionally, obtaining the object recommendation result based on the model input features corresponding to each moment when the target account engages in the object behavior and using the temporal recall model includes: encoding the model input features corresponding to each moment when the target account engages in the object behavior to obtain a vector representation of the target account at each moment when the object behavior engages; and determining the loss function as:

[0012] Among them, τ represents the model loss value, h s represents a merged vector representation of any moment when the object behavior occurs within the preset time period, is the vector representation of the next moment after any one of the moments when the object behavior occurs, is the merged vector representation of any moment after any moment at which the object behavior occurs, and E is a preset similarity threshold, wherein the merged vector representation is obtained by merging the vector representations of all moments before the corresponding moment within the preset time period; based on the vector representation of the target account at each moment at which the object behavior occurs, and the loss function, the temporal recall model is used to obtain the object recommendation result.

[0013] Optionally, when the account type is the second type of account, the object recommendation result of the target account is obtained based on the object details and the hot word recall model, including: based on the object details, using the word frequency-inverse document frequency method to obtain multiple object categories triggered by the target account within the preset time period, and the first weights corresponding to the multiple object categories respectively; determining the behavior paths corresponding to the multiple object categories, the behavior weights and the number of behaviors corresponding to the behavior paths, wherein the behavior paths are used to indicate the changes in the object behavior of the target account; according to the multiple object categories, the first weights corresponding to the multiple object categories, and the behavior weights and the number of behaviors corresponding to the multiple object categories, using the Newton's law of cooling model, obtain the second weights corresponding to the multiple object categories respectively; based on the second weights corresponding to the multiple object categories respectively, determine the object recommendation result of the target account.

[0014] Optionally, the obtaining, based on the multiple object categories, the first weights respectively corresponding to the multiple object categories, and the behavior weights and the number of behaviors respectively corresponding to the multiple object categories, and using the Newton's law of cooling model, respectively corresponding to the multiple object categories, includes: obtaining, based on the multiple object categories, the first weights respectively corresponding to the multiple object categories, and the behavior weights and the number of behaviors respectively corresponding to the multiple object categories, and using the Newton's law of cooling model, the second weight corresponding to any one of the multiple object categories in the following manner:

[0015] Among them, ε j is the second weight corresponding to any one of the object categories, θ b is the behavior weight corresponding to any one of the object categories, ν is the number of behaviors corresponding to any one of the object categories, j is the identifier corresponding to any one of the object categories, ω tf is the first weight corresponding to any one of the object categories, k is a preset time attenuation coefficient, t0 is the occurrence time of the behavior path corresponding to any one of the object categories, and t is the current time.

[0016] Optionally, obtaining the hot search words of the target account based on the object recommendation result includes: obtaining the object title in the e-commerce platform; performing named entity recognition on the object title to obtain an entity tag mapping table, wherein the entity tag mapping table includes the entity tag corresponding to the object title and the entity value corresponding to the entity tag; determining the target entity tag corresponding to the object recommendation result from the entity tag mapping table; and determining the hot search words of the target account based on the object recommendation result and the target entity tag.

[0017] According to another aspect of an embodiment of the present application, a search hot word determination device is also provided, including: a first acquisition module, used to obtain object behavior data of a target account within a preset time period, wherein the object behavior data includes: the number of object behaviors of the target account within the preset time period, the time when the object behavior occurs and the trigger type of the object behavior, and the preset time period is a period of predetermined length before the current time; a merging module, used to merge the object information and the object behavior data according to the object identifier to obtain object details, wherein the object details are used to indicate the correspondence between the object behavior of the target account and the object information; a first determination module, used to determine the account type of the target account and the hot word recall model corresponding to the account type based on the number of object behaviors of the target account within the preset time period; a second acquisition module, used to obtain the object recommendation result of the target account based on the object details and the hot word recall model; a third acquisition module, used to obtain the search hot word of the target account based on the object recommendation result.

[0018] According to another aspect of an embodiment of the present application, an electronic device is also provided, comprising one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement any one of the search hot word determination methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the conventional technology, the following briefly introduces the drawings required for use in the embodiments or the conventional technology descriptions. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the disclosed drawings without any creative work.

[0020] FIG1 is a schematic diagram of a method for determining a hot search word according to some embodiments of the present application;

[0021] FIG2 is a schematic diagram of an optional method for determining hot search words according to some embodiments of the present application;

[0022] FIG3 is a schematic diagram of another optional method for determining hot search words according to some embodiments of the present application;

[0023] FIG4 is a schematic diagram of another optional method for determining hot search words according to some embodiments of the present application;

[0024] FIG5 is a schematic diagram of a device for determining a hot search word according to some embodiments of the present application. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only 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.

[0026] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in a sequence other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0027] According to an embodiment of the present application, an embodiment of a method for determining search hot words is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0028] FIG1 is a flow chart of a method for determining a hot search term according to an embodiment of the present application. As shown in FIG1 , the method includes the following steps:

[0029] Step S102, obtaining the object behavior data of the target account within a preset time period, wherein the object behavior data includes: the number of object behaviors occurring in the target account within the preset time period, the time when the object behavior occurs, and the trigger type that triggers the object behavior. The preset time period is a period of a predetermined length before the current time.

[0030] Optionally, the object may be a product, and the corresponding object behavior is a product behavior. The trigger types of product behaviors may include, but are not limited to, user clicks, views, orders, searches, and other behaviors. Each user corresponds to an account. The corresponding object behavior data is product behavior data, including the number of product behaviors that occurred within a preset time period, the time at which the product behavior occurred, and the trigger type that triggered the product behavior. The preset time period is a continuous period of a predetermined length before the current time.

[0031] Step S104 : merging the object information and the object behavior data according to the object identifier to obtain object details, wherein the object details are used to indicate the corresponding relationship between the object behavior and the object information of the target account.

[0032] Optionally, the above-mentioned object information is presented in the form of a table. Taking the object information table as an example, the product information table includes the product number, product category, and product price corresponding to each product included in the e-commerce platform. The product category can be one or more levels. Different category levels are used to indicate the degree of subdivision of the product category. The higher the category level, the finer the product category division. For example, taking electronic products as an example: the first-level category is electronic products; the second-level categories are mobile phones, televisions, computers, cameras, etc.; the third-level categories are smartphones, tablet phones, LCD TVs, ultra-high-definition TVs, desktop computers, laptops, SLR cameras, mirrorless cameras, etc. For another example, taking clothing as an example: the first-level category is clothing; the second-level categories are men's clothing, women's clothing, children's clothing, etc.; the third-level categories are men's jackets, women's skirts, children's tops, etc. This hierarchical classification method can make it easier to manage and search for products.

[0033] Optionally, still taking the object as a product as an example, the corresponding object identifier is the product identifier, and the object details are the product details. Through data embedding, the user click access click, view pageview, and search library data are uniformly extracted to obtain the user click, view, order, search and other product behavior data within the preset time period, and pre-process the product behavior data. First, convert the time into a timestamp, remove redundant data that users repeatedly click and view in a short period of time, remove dirty data with empty product content, delete non-core channel data, sort the click, view and other behavior data according to the interaction time series, and merge the product information table and the product behavior data table according to the product identifier (such as the product number) using the Cartesian product method to obtain the product details of the account that has the product behavior. The product details can be presented in the form of a product details table. Specifically, the Cartesian product method is to match the elements in the product information table and the product behavior data table one by one according to the product identifier to generate a new set containing all possible combinations as the product details table.

[0034] Step S106 , based on the number of times the target account performs the target behavior within a preset time period, the account type of the target account and the hot word recall model corresponding to the account type are determined.

[0035] Optionally, the number of times a target account performs an object behavior within a preset time period can be used to characterize the target account's activity on the e-commerce platform. Based on the number of times a target account performs an object behavior within a preset time period, the account can be divided into three types: if the number of times a target account performs an object behavior within a preset time period is large and greater than a first preset number, the target account is determined to be a first-type account, and the first-type account can be understood as an active account; if the number of times a target account performs an object behavior within a preset time period is moderate and between the first preset number and a second preset number (the second preset number is less than the first preset number), the target account is determined to be a second-type account, and the second-type account can be understood as an inactive account; if the number of times a target account performs an object behavior within a preset time period is small and less than the second preset number, the target account is determined to be a third-type account, and the third-type account can be understood as a visitor account that does not frequently visit the e-commerce platform.

[0036] It should be noted that different types of accounts have different behavioral characteristics. If the same method is used to select hot words, it will not be able to reflect the characteristics of each type of account, which may easily lead to inaccurate hot word recommendations. Based on this, different types of hot word recall models are selected for different types of accounts to recommend hot words, so as to improve the accuracy of hot word recommendations and thus enhance the user experience.

[0037] In an optional embodiment, the account type of the target account and the hot word recall model corresponding to the account type are determined based on the number of times the target account performs the object behavior within a preset time period, including: when the number of times the target account performs the object behavior within the preset time period is greater than a first preset number, the account type is determined to be a first type of account, and the corresponding hot word recall model is a time series recall model; when the number of times the target account performs the object behavior within the preset time period is less than or equal to the first preset number and greater than a second preset number, the account type is determined to be a second type of account, and the corresponding hot word recall model is a Newton's cooling law model.

[0038] Optionally, for the first category of accounts with high activity, a time-series recall Core model is constructed as the hot word recall model for hot word recommendation; for the second category of accounts with moderate activity, Newton's law of cooling is used as the hot word recall model for hot word recommendation. Through the above methods, the corresponding hot word recall model is selected according to account activity for hot word recommendation, and a Core model and cooling model diversion strategy based on account activity is constructed to achieve personalized hot word recommendation.

[0039] In an optional embodiment, the method further includes: when the number of object behaviors that occur in the target account within a preset time period is less than or equal to a second preset number, obtaining the search record hot words corresponding to the accounts included in the e-commerce platform, wherein the search record hot words are determined based on the historical search records of the accounts included in the e-commerce platform; based on the search record hot words corresponding to the accounts included in the e-commerce platform, determining the search hot words of the target account.

[0040] Optionally, for the third type of accounts with lower activity intensity, since this type of accounts has a low login frequency for the e-commerce platform and it is impossible to know the access preferences of this type of accounts, it is possible to directly recommend hot search terms based on the historical search records of all accounts on the e-commerce platform within a certain period of time. This allows this type of accounts to know the hot search terms of all accounts on the current platform to assist the accounts in making corresponding search decisions.

[0041] Step S108: Based on the object details, a hot word recall model is used to obtain object recommendation results for the target account.

[0042] Alternatively, using products as an example, a hot word recall model can be used based on product details to determine product recommendations for the target account. The product recommendation results will at least include the recommended product category, such as the secondary category of the recommended product.

[0043] In an optional embodiment, when the account type is a first-category account, based on the object details, a hot word recall model is used to obtain the object recommendation result of the target account, including: based on the object details, obtaining the model input features corresponding to each moment when the object behavior occurs in the target account within a preset time period, wherein the model input features include: feature triples, and at least one of the following: object category, object price and sales volume, object marketing score, object title, target account information, the feature triples include target account ID, object ID and timestamp information; based on the model input features corresponding to each moment when the object behavior occurs in the target account, a temporal recall model is used to obtain the object recommendation result of the target account.

[0044] Optionally, when the target account type is a first-class account, a hot word recall model of the Core model based on the active user behavior characteristics is constructed based on the product details table, and the account interest is predicted through the model reasoning ability, so as to recall a set of product identifiers consisting of the identifiers of the number of products booked before the account interest (such as 20) as the product recommendation result, wherein the identifiers of the 20 products can be the identifiers corresponding to the secondary categories of the 20 products. Specifically, first extract the account identifier users-product identifier items-timestamp timestamp feature triples according to the product details table, and construct feature items such as product category, product price and sales volume, product marketing score, product title, account information, etc. as model input features. Encode and map the numbers in the above model input features, split and encode the text features, and uniformly convert them into digital feature items. Digitally encode the product features corresponding to the account through the encoding Embedding layer to obtain the representation vector item-emb. Build the core Core temporal recall model, uniformly obtain the identification vector item-emb related features through the Transformer Encoder layer, shift the account temporal window, merge the temporal interaction features under the current window with the next temporal interaction features, and predict the item-emb features of all representation vectors of subsequent temporal accounts. Based on the item-emb features of all representation vectors of subsequent temporal accounts, complete the reasoning capability of the Core model. By inputting the account identification id and the set of product identification ids of the account's recent behavior, it is possible to infer the set of product identification ids that the account is more interested in in the current temporal sequence. Based on this set of product identifications, the product recommendation results can be obtained. Based on this temporal recall model, by analyzing the user's historical behavior sequence, the interest evolution law of the target account can be learned, thereby providing personalized recommendations.

[0045] In an optional embodiment, based on the model input features corresponding to each time when the target account performs an object behavior, a temporal recall model is used to obtain an object recommendation result, including: encoding the model input features corresponding to each time when the target account performs an object behavior to obtain a vector representation of the target account at each time when the object behavior occurs; and determining the loss function as follows:

[0046] Among them, τ represents the model loss value, h s It represents the merged vector representation of any moment when the object behavior occurs within the preset time period. is the vector representation of the next moment after any object behavior occurs, is the merged vector representation of any moment after any moment when the object behavior occurs, and E is the preset similarity threshold, where the merged vector representation is obtained by merging the vector representations of all moments before the corresponding moment within the preset time period; based on the vector representation of the target account at each moment when the object behavior occurs and the loss function, the temporal recall model is used to obtain the object recommendation result.

[0047] Optionally, to improve the associative power of the Time Series Recall Core model, a weight threshold can be added to the regular loss function to calculate the similarity between the current moment and the subsequent moment, reducing overfitting caused by overly similar item-emb vector representations. Setting the similarity threshold to 0.75 prevents over-learning of similarities and enhances model generalization.

[0048] In an optional embodiment, when the account type is a second-category account, a hot word recall model is used based on the object details to obtain object recommendation results for the target account, including: based on the object details, a word frequency-inverse document frequency method is used to obtain multiple object categories triggered by the target account within a preset time period, and first weights corresponding to the multiple object categories respectively; determining the behavior paths corresponding to the multiple object categories, the behavior weights and the number of behaviors corresponding to the behavior paths, wherein the behavior paths are used to indicate changes in the object behavior of the target account; based on the multiple object categories, the first weights corresponding to the multiple object categories, and the behavior weights and the number of behaviors corresponding to the multiple object categories, a Newton's law of cooling model is used to obtain second weights corresponding to the multiple object categories respectively; based on the second weights corresponding to the multiple object categories respectively, the object recommendation results for the target account are determined.

[0049] Optionally, building a Newton's cooling law model based on the behavioral characteristics of inactive accounts can effectively address the cold start problem for inactive accounts. Specifically, based on the behavioral paths in the product details table of the target account, different behavioral weights are assigned to different behavioral paths. This yields behavioral paths and behavioral weights corresponding to multiple product categories. Table 1 shows an example of some behavioral path-weight mappings.

[0050] Table 1

[0051] Further, secondary category information for products triggered by the target account within a preset time period is obtained. The weight information of the account's behavior on the secondary product categories is calculated based on the term frequency minus the inverse document frequency (TfIdf), resulting in first weights corresponding to each of the multiple product categories (i.e., multiple secondary product categories). Based on the multiple product categories, the first weights corresponding to each of the multiple product categories, and the behavior weights and number of behaviors corresponding to each of the multiple product categories, a Newton's law of cooling model is used to calculate the account's temporal relevance to the products, resulting in second weights corresponding to each of the multiple product categories. These second weights can serve as the target account's interest weights in the product categories. Based on the calculation of the second weights, the second weights corresponding to the multiple product categories are sorted. Based on the sorting results, the target account's recently visited secondary product categories are screened for secondary categories of greater interest, resulting in a set of recommended product IDs as product recommendation results. This Newton's law of cooling model can be used to predict popular search terms or popular content that the account may be interested in at the current moment based on the account's historical search or click behavior, allowing personalized recommendations for inactive accounts searching for hot search terms.

[0052] In an optional embodiment, a Newton's law of cooling model is used to obtain second weights corresponding to the plurality of object categories based on a plurality of object categories, first weights corresponding to the plurality of object categories, and behavior weights and behavior counts corresponding to the plurality of object categories. The method includes: obtaining the second weight corresponding to any one of the plurality of object categories based on the plurality of object categories, first weights corresponding to the plurality of object categories, and behavior weights and behavior counts corresponding to the plurality of object categories using the Newton's law of cooling model in the following manner:

[0053] Among them, ε j is the second weight corresponding to any object category, θ b is the behavior weight corresponding to any object category, ν is the number of behaviors corresponding to any object category, j is the identifier corresponding to any object category, ω tf is the first weight corresponding to any object category, k is the preset time attenuation coefficient, t0 is the occurrence time of the behavior path corresponding to any object category, and t is the current time.

[0054] Optional, This represents the time decay from the time the account behavior occurred to the current time. Time decay can be set to a daily basis to reduce the rapid cooling of interest due to account inactivity. The time decay coefficient can be set to 0.125. The above formula can be used to calculate the target account's interest weight for a product category based on time series.

[0055] Step S110: Obtain the target account's hot search terms based on the object recommendation results.

[0056] Optionally, still taking the object as a product as an example, the product recommendation results record a product identification set consisting of the identifications of the products that the target account has been interested in recently (within a preset time period). Based on the product identification set, a product hot word set can be recommended and returned in real time, realizing the real-time diversion and recommendation capability of hot words.

[0057] In an optional embodiment, based on the object recommendation result, the search hot words of the target account are obtained, including: obtaining the object title in the e-commerce platform; performing named entity recognition on the object title to obtain an entity tag mapping table, wherein the entity tag mapping table includes an entity tag corresponding to the object title and an entity value corresponding to the entity tag; determining the target entity tag corresponding to the object recommendation result from the entity tag mapping table; and determining the search hot words of the target account based on the object recommendation result and the target entity tag.

[0058] Optionally, the e-commerce data storage within the e-commerce platform uses the big data warehouse capability to pull e-commerce details to obtain relevant information such as product identification, product category, and product price. By clicking on the e-commerce account behavior, viewing the pageview, and searching the search library, relevant information such as account identification, account behavior events, behavior path, and behavior time is obtained. The existing e-commerce product title entity knowledge base is sorted out, and a w2ner named entity extraction model is constructed and trained. Some examples of product entity label types are shown in Table 2 below. Then, the model reasoning capability is used to extract unlabeled product entities, and by building a product vocabulary and integrating regular matching capabilities into the reasoning link, the model's ability to extract entities such as product brand words and specification types is compensated, the accuracy of product labels is improved, and the cost of manual labeling is reduced, resulting in an entity label mapping table corresponding to the product title. The form of the entity label mapping table is shown in Table 2 below, which is used to indicate the mapping relationship between product titles and product labels. Based on the product details table and the product entity tag mapping table, an entity tag extraction strategy is used. The strategy can use the modified tags in Table 2 of the product entity tags to splice the product entity as the recall word set. The real-time account data input account triggers the product identification ID set (i.e., the product recommendation result) nearly a predetermined number of times (such as nearly 50 times). The product hot word set can be recommended in real time and returned, realizing the real-time diversion recommendation capability of hot words.

[0059] Table 2

[0060] Through the above steps S102 to S110, the purpose of accurately recommending hot words by selecting the corresponding hot word recall model according to different account types can be achieved, thereby realizing the technical effect of improving the accuracy of search hot word recommendation results, and then improving user experience, thereby solving the technical problem of low prediction accuracy and poor applicability in the related technology of using a single hot word recall model to predict search hot word recommendation results for all accounts.

[0061] Based on the above embodiments and optional embodiments, the present application proposes an optional implementation manner. FIG2 is a flowchart of an optional method for determining a hot search word according to an embodiment of the present application. As shown in FIG2 , the method includes:

[0062] Step S1: Product tag extraction. Obtain product titles from the e-commerce platform; use the pre-built w2ner named entity extraction model to perform named entity recognition on the product titles to obtain an entity tag mapping table, where the entity tag mapping table is used to indicate the mapping relationship between product titles and product tags.

[0063] Step S2, account tracking data extraction and cleaning. Through data tracking, the account click access click, view pageview, search library data are uniformly extracted to obtain the account click, view, order, search and other product behavior data within a preset time period. The data is pre-processed. First, the time is converted into a timestamp, and redundant data of repeated clicks and views of the account in a short period of time is eliminated. The dirty data with empty product content is eliminated, and non-core channel data is deleted. The click and view behavior data are sorted according to the interaction time series, and the product information table and the product behavior data table are merged by Cartesian product according to the product identifier to obtain the product details of the account's product behavior. The product details can be presented in the form of a product details table.

[0064] Step S3: Constructing a hot word recall model based on the Core temporal recall model of active account behavior characteristics. For the product details table provided above, a hot word recall model based on the Core model of active account behavior characteristics is constructed. Account interests are predicted through model reasoning capabilities, thereby recalling the top 20 product ID sets of account interests as product recommendation results. FIG3 is a flowchart of another optional search hot word determination method according to an embodiment of the present application. As shown in FIG3, the specific process is as follows:

[0065] Step S31: Extract the account ID users-product ID items-timestamp feature triples according to the product details table in step S2, and construct feature items such as product category, product price and sales volume, product marketing score, product title, and account information as model input features.

[0066] In step S32, the numbers in the input features of the above model are encoded and mapped, the text features are split and encoded, and uniformly converted into digital feature items. The product features corresponding to the account are digitally encoded through the encoding Embedding layer to obtain the vector representation item-emb.

[0067] Step S33: Build the core Core temporal recall model, uniformly obtain the item-emb related features of the identification vector through the Transformer Encoder layer, shift the account temporal window, merge the temporal interaction features under the current window with the next temporal interaction features, and predict the item-emb features of all vector representations of subsequent temporal accounts.

[0068] In step S34, in order to improve the associative ability of the Core model, a weight threshold is set on the conventional loss function to calculate the similarity between the current moment and the subsequent moment, thereby reducing overfitting caused by the vector representation of item-emb being too similar.

[0069] Step S35 completes the reasoning capability of the Core model. By inputting the account ID and the product ID set of the account's recent behavior, the product ID set that the account is more interested in in the current time sequence can be inferred. Based on this product ID set, product recommendation results can be obtained.

[0070] Step S4: Constructing a hot word recall model based on the Newton's cooling law model of the inactive account behavior characteristics. Constructing a Newton's cooling law model based on the inactive account behavior characteristics can effectively solve the cold start problem of inactive accounts. FIG4 is a flowchart of another optional search hot word determination method according to an embodiment of the present application, as shown in FIG2. The specific process is as follows:

[0071] In step S41 , according to the behavior path of the product details table of the target account in step S2 , different behavior weights are set for different behavior paths to obtain behavior paths and behavior weights corresponding to multiple product categories.

[0072] Step S42: Obtain the secondary category information of the product triggered by the target account within a preset time period, calculate the weight information of the account behavior on the secondary category of the product based on the word frequency-inverse document frequency TfIdf, and obtain the first weights corresponding to multiple product categories (i.e., multiple secondary categories of products).

[0073] In step S43, based on the multiple product categories, the first weights corresponding to each of the multiple product categories, and the behavior weights and behavior counts corresponding to each of the multiple product categories, a Newton's law of cooling model is used to implement the time-series product relevance of the account. Second weights corresponding to each of the multiple product categories are obtained. These second weights can serve as the target account's interest weights in the product categories. Based on the calculation of the second weights, the second weights corresponding to the multiple product categories are sorted. Based on the sorting results, the target account's recently visited secondary product categories are screened for secondary categories of greater interest to the target account, resulting in a set of recommended product IDs as the product recommendation results.

[0074] Step S5: Real-time account behavior is input to retrieve the account's corresponding product interest ranking set. Steps S3 and S4 are used to retrieve product interest for accounts with different levels of activity. Accounts are categorized by activity level into active, inactive, and visitor categories. Active accounts are retrieved using the Core model using their account IDs, while inactive accounts are retrieved using the Cooling Model. For visitor accounts, hot word recall is performed using the search history of all accounts.

[0075] Step S6: Return the account's search hot words based on the product tags. Based on the product details table and the product entity tag mapping table, an entity tag extraction strategy is used. The strategy can use the modified tags in the product entity tag and the product entity as the recall word set. The account's real-time data input is the set of product IDs triggered by the account nearly 50 times (i.e., the product recommendation results). The product hot word set can be recommended in real time and returned, realizing the real-time hot word diversion recommendation capability.

[0076] In the embodiment of the present application, the content of the present application is the hot word recommendation capability for e-commerce business scenarios, and proposes 1) extracting a hot word set from the product title in the e-commerce scenario by constructing a named entity model. The hot word construction strategy adopts a strategy of splicing modifiers and product entity words to obtain a hot word recall set corresponding to the product; 2) dividing active accounts according to the number of account behavior intervals, constructing a Core model and cooling model diversion strategy based on account activity, and optimizing the Core model loss function to prevent overfitting; 3) integrating marketing strategy weights into the Core model and cooling model respectively, and adjusting the sorting and display of recommended hot words according to the marketing weights; 4) pulling the recent behavior of the account in real time, inputting the diversion dual model, and recalling the recommended hot word results in real time.

[0077] It should be noted that the embodiment of the present application proposes a method for diverting hot word recommendations in e-commerce scenarios based on real-time account behavior. By extracting a named entity model, the cost of manual annotation of e-commerce product data is reduced, and annotation efficiency is improved. By combining the historical behavior of accounts and the characteristics of real-time data, a fusion diversion strategy of the two models is constructed to improve the concurrent carrying capacity and personalization of recommendations, while solving the cold start problem of inactive accounts due to effective clicks. By integrating marketing strategy weight information, it helps to improve the effectiveness of e-commerce marketing, be more in line with the business strategy of the e-commerce platform, and enhance the account experience.

[0078] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0079] In this embodiment, a device for determining a hot search term is also provided. This device is used to implement the above-mentioned embodiments, and the details that have been described will not be repeated here. As used below, the terms "module" and "device" can refer to a combination of software and / or hardware that implements a predetermined function. The devices described in the following embodiments can be implemented in software, hardware, or a combination of software and hardware. Implementation is also possible and contemplated.

[0080] According to an embodiment of the present application, a device embodiment for implementing the above-mentioned method for determining hot search words is also provided. FIG5 is a structural diagram of a device for determining hot search words according to an embodiment of the present application. As shown in FIG5 , the device for determining hot search words includes: a first acquisition module 500, a merging module 502, a first determination module 504, a second acquisition module 506, and a third acquisition module 508, wherein:

[0081] A first acquisition module 500 is configured to acquire object behavior data of a target account within a preset time period, wherein the object behavior data includes: the number of times the target account has performed the object behavior within the preset time period, the time when the object behavior occurred, and the trigger type that triggered the object behavior. The preset time period is a period of a predetermined length before the current time.

[0082] A merging module 502, connected to the first acquisition module 500, is configured to merge the object information and the object behavior data according to the object identifier to obtain object details, wherein the object details are used to indicate the correspondence between the object behavior and the object information of the target account;

[0083] A first determination module 504, connected to the merging module 502, is configured to determine the account type of the target account and a hot word recall model corresponding to the account type based on the number of times the target account has performed the target behavior within a preset time period;

[0084] The second acquisition module 506 is connected to the first determination module 504 and is used to obtain the object recommendation result of the target account based on the object details and using the hot word recall model;

[0085] The third acquisition module 508 is connected to the second acquisition module 506 and is used to obtain the search hot words of the target account based on the object recommendation results.

[0086] In an embodiment of the present application, a first acquisition module 500 is used to acquire the object behavior data of the target account within a preset time period, wherein the object behavior data includes: the number of object behaviors occurring in the target account within the preset time period, the time when the object behavior occurs, and the trigger type of the object behavior, wherein the preset time period is a period of predetermined length before the current time; a merging module 502 is connected to the first acquisition module 500 and is used to merge the object information and the object behavior data according to the object identifier to obtain object details, wherein the object details are used to indicate the corresponding relationship between the object behavior and the object information of the target account; a first determination module 504 is connected to the merging module 502 and is used to determine the target account based on the number of object behaviors occurring in the preset time period. The account type of the target account, and the hot word recall model corresponding to the account type; the second acquisition module 506, connected to the first determination module 504, is used to obtain the object recommendation result of the target account based on the object details and the hot word recall model; the third acquisition module 508, connected to the second acquisition module 506, is used to obtain the search hot words of the target account based on the object recommendation result, thereby achieving the purpose of accurately recommending hot words by selecting the corresponding hot word recall model according to different account types, thereby achieving the technical effect of improving the recommendation accuracy of the search hot word recommendation results and thus improving the user experience, thereby solving the technical problem of low prediction accuracy and poor applicability in the related technology of using a single hot word recall model to predict the search hot word recommendation results for all accounts.

[0087] It should be noted that the above modules can be implemented by software or hardware. For example, for the latter, it can be implemented in the following ways: the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0088] It should be noted that the first acquisition module 500, merging module 502, first determination module 504, second acquisition module 506, and third acquisition module 508 correspond to steps S102 to S110 in the embodiment. The examples and application scenarios implemented by these modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above embodiment. It should be noted that the above modules, as part of the device, can be run on a computer terminal.

[0089] It should be noted that the optional implementation methods of this embodiment can be found in the relevant descriptions in the embodiments and will not be repeated here.

[0090] The above-mentioned search hot word determination device can also include a processor and a memory. The above-mentioned first acquisition module 500, merging module 502, first determination module 504, second acquisition module 506, third acquisition module 508, etc. are all stored in the memory as program modules, and the processor executes the above-mentioned program modules stored in the memory to realize corresponding functions.

[0091] The processor includes a core, which retrieves corresponding program modules from memory. There can be one or more cores. Memory may include non-permanent memory in a computer-readable medium, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory includes at least one memory chip.

[0092] According to an embodiment of the present application, an embodiment of a non-volatile storage medium is also provided. Optionally, in this embodiment, the non-volatile storage medium includes a stored program, wherein when the program is executed, the device where the non-volatile storage medium is located is controlled to execute any of the above-mentioned search hot word determination methods.

[0093] Optionally, in this embodiment, the non-volatile storage medium may be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group, and the non-volatile storage medium includes a stored program.

[0094] Optionally, when the program is running, the device where the non-volatile storage medium is located is controlled to perform the following functions: obtaining object behavior data of the target account within a preset time period, wherein the object behavior data includes: the number of object behaviors occurring in the target account within the preset time period, the time when the object behavior occurs and the trigger type of the object behavior, and the preset time period is a period of predetermined length before the current time; merging the object information and the object behavior data according to the object identifier to obtain object details, wherein the object details are used to indicate the correspondence between the object behavior and the object information of the target account; determining the account type of the target account and the hot word recall model corresponding to the account type based on the number of object behaviors occurring in the target account within the preset time period; obtaining object recommendation results for the target account based on the object details; and obtaining search hot words for the target account based on the object recommendation results.

[0095] According to an embodiment of the present application, an embodiment of a processor is further provided. Optionally, in this embodiment, the processor is used to run a program, wherein the program executes any of the above-mentioned methods for determining hot search words when the program is run.

[0096] According to an embodiment of the present application, an embodiment of a computer program product is also provided, which, when executed on a data processing device, is suitable for executing a program that initializes any one of the above-mentioned search hot word determination method steps.

[0097] Optionally, the above-mentioned computer program product, when executed on a data processing device, is suitable for executing a program initialized with the following method steps: obtaining object behavior data of a target account within a preset time period, wherein the object behavior data includes: the number of object behaviors occurring in the target account within the preset time period, the time when the object behavior occurs and the trigger type of the object behavior, and the preset time period is a period of predetermined length before the current time; merging the object information and the object behavior data according to the object identifier to obtain object details, wherein the object details are used to indicate the correspondence between the object behavior and the object information of the target account; determining the account type of the target account and the hot word recall model corresponding to the account type based on the number of object behaviors occurring in the target account within the preset time period; obtaining object recommendation results for the target account based on the object details; and obtaining search hot words for the target account based on the object recommendation results.

[0098] An embodiment of the present application provides an electronic device, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, the following steps are implemented: obtaining object behavior data of a target account within a preset time period, wherein the object behavior data includes: the number of times the target account has had object behaviors within the preset time period, the moment when the object behavior occurred, and the trigger type for triggering the object behavior, wherein the preset time period is a period of a predetermined length before the current moment; merging the object information and the object behavior data according to the object identifier to obtain object details, wherein the object details are used to indicate the correspondence between the object behavior and the object information of the target account; determining the account type of the target account and a hot word recall model corresponding to the account type based on the number of times the object behavior has occurred within the preset time period; obtaining object recommendation results for the target account based on the object details and the hot word recall model; and obtaining search hot words for the target account based on the object recommendation results.

[0099] The above sequence of the embodiments of the present application is for description only and does not represent the superiority or inferiority of the embodiments.

[0100] In the above embodiments of the present application, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0101] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only exemplary. For example, the division of the above modules can be a logical function division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, modules or indirect coupling or communication connection of modules, which can be electrical or other forms.

[0102] The modules described above as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed across multiple modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment.

[0103] In addition, the functional modules in the various embodiments of the present application may be integrated into a processing module, or each module may exist physically separately, or two or more modules may be integrated into a single module. The above-mentioned integrated modules may be implemented in the form of hardware or software functional modules.

[0104] If the above-mentioned integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can be stored in a computer-readable non-volatile storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a non-volatile storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned non-volatile storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program code.

[0105] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0106] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for determining a hot search word, comprising: Obtaining object behavior data of a target account within a preset time period, wherein the object behavior data includes: the number of times the target account has an object behavior within the preset time period, the time when the object behavior occurs, and the trigger type of the object behavior, wherein the preset time period is a time period of a predetermined length before the current time; According to the object identifier, the object information and the object behavior data are combined to obtain object details, wherein the object details are used to indicate the corresponding relationship between the object behavior of the target account and the object information; Determine the account type of the target account and the hot word recall model corresponding to the account type based on the number of times the target account has the object behavior within the preset time period; Based on the object details, the hot word recall model is used to obtain the object recommendation result of the target account; Based on the object recommendation result, the hot search words of the target account are obtained.

2. The method according to claim 1, wherein: The determining the account type of the target account and the hot word recall model corresponding to the account type based on the number of times the target account has the object behavior within the preset time period includes: When the number of times the target account has the object behavior within the preset time period is greater than the first preset number, the account type is determined to be a first type account, and the corresponding hot word recall model is a time series recall model; When the number of times the target account performs the object behavior within the preset time period is less than or equal to the first preset number and greater than the second preset number, the account type is determined to be a second-type account, and the corresponding hot word recall model is the Newton's cooling law model, wherein the second preset number is less than the first number.

3. The method according to claim 2, wherein: In the case where the account type is the first type of account, the object recommendation result of the target account is obtained based on the object details and using the hot word recall model, including: Based on the object details, a model input feature corresponding to each moment when the target account has an object behavior within the preset time period is obtained, wherein the model input feature includes: a feature triplet, and at least one of the following: object category, object price and sales volume, object marketing score, object title, and target account information, wherein the feature triplet includes a target account identifier, an object identifier, and timestamp information; Based on the model input features corresponding to the target account at each moment when the object behavior occurs, the temporal recall model is used to obtain the object recommendation result of the target account.

4. The method according to claim 3, wherein: The object recommendation result is obtained by using the temporal recall model based on the model input feature corresponding to each time when the target account has the object behavior, including: Encoding the model input features corresponding to each time when the target account has the object behavior, to obtain a vector representation of the target account at each time when the object behavior has occurred; The loss function is determined as: Among them, τ represents the model loss value, h s represents a merged vector representation of any moment when the object behavior occurs within the preset time period, is the vector representation of the next moment after any moment when the object behavior occurs, h vi is the merged vector representation of any moment after any moment when the object behavior occurs, and E is a preset similarity threshold, wherein the merged vector representation is obtained by merging the vector representations of all moments before the corresponding moment within the preset time period; Based on the vector representation of the target account at each moment when the object behavior occurs and the loss function, the temporal recall model is adopted to obtain the object recommendation result.

5. The method according to claim 2, wherein: In the case where the account type is the second type of account, the object recommendation result of the target account is obtained based on the object details and using the hot word recall model, including: Based on the object details, a word frequency-inverse document frequency method is used to obtain a plurality of object categories triggered by the target account within the preset time period, and first weights respectively corresponding to the plurality of object categories; Determine the behavior paths corresponding to the multiple object categories, the behavior weights and behavior times corresponding to the behavior paths, wherein the behavior paths are used to indicate the object behavior changes of the target account; According to the multiple object categories, the first weights respectively corresponding to the multiple object categories, and the behavior weights and the behavior times respectively corresponding to the multiple object categories, the Newton's law of cooling model is used to obtain second weights respectively corresponding to the multiple object categories; The object recommendation result of the target account is determined based on the second weights respectively corresponding to the multiple object categories.

6. The method according to claim 5, wherein: The method of obtaining second weights corresponding to the plurality of object categories respectively by using the Newton's cooling law model according to the plurality of object categories, the first weights corresponding to the plurality of object categories respectively, and the behavior weights and the behavior times corresponding to the plurality of object categories respectively, comprises: According to the multiple object categories, the first weights respectively corresponding to the multiple object categories, and the behavior weights and the behavior times respectively corresponding to the multiple object categories, the second weight corresponding to any one of the multiple object categories is obtained by using the Newton's cooling law model in the following manner: Among them, ε j is the second weight corresponding to any one of the object categories, θ b is the behavior weight corresponding to any one of the object categories, ν is the number of behaviors corresponding to any one of the object categories, j is the identifier corresponding to any one of the object categories, ω tf is the first weight corresponding to any one of the object categories, k is a preset time attenuation coefficient, t0 is the occurrence time of the behavior path corresponding to any one of the object categories, and t is the current time.

7. The method according to claim 2, wherein: The method further comprises: When the number of times the target account has the object behavior within the preset time period is less than or equal to the second preset number, obtaining search record hot words corresponding to the accounts included in the e-commerce platform, wherein the search record hot words are determined based on historical search records of the accounts included in the e-commerce platform; The search hot words of the target account are determined based on the search record hot words respectively corresponding to the accounts included in the e-commerce platform.

8. The method according to any one of claims 1 to 7, wherein: The obtaining the search hot words of the target account based on the object recommendation result includes: Get the object title in the e-commerce platform; Performing named entity recognition on the object title to obtain an entity label mapping table, wherein the entity label mapping table includes an entity label corresponding to the object title and an entity value corresponding to the entity label; Determine a target entity tag corresponding to the object recommendation result from the entity tag mapping table; Based on the object recommendation result and the target entity tag, the hot search word of the target account is determined.

9. A device for determining a hot search word, comprising: A first acquisition module is used to acquire object behavior data of a target account within a preset time period, wherein the object behavior data includes: the number of times the target account has an object behavior within the preset time period, the time when the object behavior occurs, and the trigger type of the object behavior, and the preset time period is a period of a predetermined length before the current time; a merging module, configured to merge the object information and the object behavior data according to the object identifier to obtain object details, wherein the object details are used to indicate the corresponding relationship between the object behavior of the target account and the object information; A first determination module, configured to determine the account type of the target account and a hot word recall model corresponding to the account type based on the number of times the target account has the object behavior within the preset time period; A second acquisition module is used to obtain the object recommendation result of the target account based on the object details and using the hot word recall model; The third acquisition module is used to obtain the hot search words of the target account based on the object recommendation result.

10. An electronic device comprising one or more processors and a memory, wherein the memory is used to store one or more programs, wherein: When the one or more programs are executed by the one or more processors, the one or more processors are enabled to implement the steps of the method according to any one of claims 1 to 8.

Images