CN117235359A

CN117235359A - Search recommendation method, device and storage medium

Info

Publication number: CN117235359A
Application number: CN202311182611.8A
Authority: CN
Inventors: 穆瑞斌
Original assignee: Rajax Network Technology Co Ltd; Zhejiang Koubei Network Technology Co Ltd
Current assignee: Rajax Network Technology Co Ltd; Zhejiang Koubei Network Technology Co Ltd
Priority date: 2023-03-31
Filing date: 2023-03-31
Publication date: 2023-12-15
Also published as: CN116089729B; CN116089729A

Abstract

One or more embodiments of the present specification provide a search recommendation method, an electronic device, and a computer-readable storage medium. The method comprises the following steps: responding to a search page display instruction, and recalling a plurality of candidate search words; under the condition that the plurality of candidate search words comprise candidate commodity words, determining all target commodity words corresponding to commodities indicated by the candidate commodity words; if the commodity named by the target commodity word can be provided for the user in the distribution range of the user, generating a target recommendation file according to the candidate commodity word; and displaying the target recommended file in a search page so as to search for goods under the condition that the target recommended file is triggered. The implementation can also show the search words meeting the supply condition to the user in a text manner under the condition that the user does not key in the search words.

Description

Search recommendation method, device and storage medium

Technical Field

One or more embodiments of the present disclosure relate to the field of terminal technologies, and in particular, to a search recommendation method, an electronic device, and a computer readable storage medium.

Background

In the related art, the search function is an entry for a user to search for information, and is an important tie connecting the user and the information. The user can input search words in the search bar, and the electronic equipment can search data according to the search words input by the user, acquire at least one search result and display the search result to the user. However, for some users who lack the purpose in searching, the user may stay in the search page for too long, affecting the user experience.

Therefore, it is necessary to provide a search recommendation method for making a search recommendation before a user does not input a search term.

Disclosure of Invention

In view of this, one or more embodiments of the present specification provide a search recommendation method, an electronic device, and a computer-readable storage medium.

In order to achieve the above object, one or more embodiments of the present disclosure provide the following technical solutions:

according to a first aspect of one or more embodiments of the present specification, a search recommendation method is provided, including:

responding to a search page display instruction, and recalling a plurality of candidate search words;

under the condition that the plurality of candidate search words comprise candidate commodity words, determining all target commodity words corresponding to commodities indicated by the candidate commodity words;

if the commodity named by the target commodity word can be provided for the user in the distribution range of the user, generating a target recommendation file according to the candidate commodity word;

and displaying the target recommended file in a search page so as to search for goods under the condition that the target recommended file is triggered.

In this embodiment, a number of candidate search words can be recalled before the user does not type in a search word, and the search word is recommended to the user in a text manner if the user does not type in a search word. In order to improve the accuracy of commodity recommendation, all commodity names with the same meaning as the candidate commodity words but different names are searched for supply judgment, and the accuracy of supply judgment is improved, so that the user can be ensured to recommend proper and useful search words, and the condition that the same commodity is misjudged in supply and is misrecommended due to different names is avoided or reduced.

In one implementation manner, the determining all target commodity words corresponding to the commodity indicated by the candidate commodity word includes:

converting the candidate commodity words into target characterization vectors by using a pre-trained neural network model; the neural network model is used for converting different commodity names of the same commodity into the same characterization vector;

and determining that the results of the conversion by using the neural network model are all target commodity words of the target characterization vector.

In this embodiment, all commodity names corresponding to the commodities indicated by the candidate commodity words are searched by converting different commodity names of the same commodity into the neural network model of the same characterization vector.

In one implementation manner, the determining that the results of the conversion performed by using the neural network model are all target commodity words of the target token vector includes:

and determining all target commodity words corresponding to the target characterization vector according to the pre-stored corresponding relations between different commodity names and the characterization vector of the same commodity.

In this embodiment, the trained neural network model is used in advance to obtain the correspondence between different commodity names and the characterization vectors of the same commodity, so as to be used for quickly searching all target commodity words corresponding to the target characterization vectors.

In one implementation, the neural network model includes an embedded layer and an encoder; the target characterization vector is obtained by converting the candidate commodity words into embedded vectors through the embedding layer and mapping the embedded vectors from a character vector space to a numerical vector space through the encoder.

In one implementation, training the neural network model includes:

obtaining a plurality of binary group samples, wherein one part of the binary group samples comprise two positive samples, and the other part of the binary group samples comprise one positive sample and one negative sample; the two positive samples comprise different commodity names of the same commodity, and the negative sample and the positive sample belong to commodity names of different commodities;

inputting the binary group samples into a neural network model to be trained with two branches, processing one of the binary group samples by each branch to obtain two characterization vectors corresponding to the binary group samples output by the two branches, and training the neural network model to be trained by taking the distance between the characterization vectors respectively corresponding to different commodity names of the same commodity as an optimization target and/or maximizing the distance between the characterization vectors respectively corresponding to the commodity names of different commodities;

Wherein the weights of the two branches are shared; the trained neural network model includes at least one of the branches.

In this embodiment, the neural network model is trained by comparing the learning mode with the characterization learning mode, which is beneficial to improving the prediction performance of the model.

acquiring all target commodity words from a commodity name list of commodities indicated by the candidate commodity words according to the candidate commodity words; the commodity name list comprises different commodity names corresponding to the same commodity.

In this embodiment, quick determination of all commodity names corresponding to commodities indicated by the candidate commodity word is achieved through a prestored commodity name table.

In one implementation, before generating the target recommendation document according to the candidate commodity word if the commodity named by the target commodity word can be provided for the user in the distribution range of the user, the method further comprises:

determining a store containing the commodity named by the target commodity word, and acquiring position information of the store;

based on the location information of the store and the location information of the user, it is determined whether or not the user can be provided with the product named by the target product word within the distribution range of the user.

In this embodiment, the supply judgment is realized based on the difference between the position information of the store containing the commodity named by the target commodity word and the position information of the user.

In one implementation, the method further comprises:

acquiring position information of a store indicated by the candidate store words under the condition that the candidate search words comprise the candidate store words;

determining whether commodities of the stores indicated by the candidate store words can be provided for the user in the distribution range of the user according to the position information of the stores indicated by the candidate store words and the position information of the user;

if yes, generating a target recommendation document according to the candidate commodity words.

In this embodiment, supply judgment of the provided candidate shop word is realized, so that the shop can be recommended to the user in a text manner under the condition that the user does not enter the search word.

In one implementation, the generating the target recommendation document according to the candidate commodity word includes:

and if the candidate commodity word is exposed and clicked in the history display process or the candidate commodity word is not exposed in the history display process, generating a target recommendation file according to the candidate commodity word.

In this embodiment, the exposed and clicked candidate search terms can satisfy the preference of the user; unexposed candidate search terms help explore and mine user interest.

dividing the plurality of candidate commodity words into at least two parts under the condition that the plurality of candidate commodity words are provided;

for a part of candidate commodity words, generating a target recommended document according to the candidate commodity words and the first candidate document template; the first candidate document template is generated according to the history recommended document which is exposed and clicked in the history display process;

for another part of candidate commodity words, generating a target recommended document according to the candidate commodity words and the second candidate document template; the second candidate document template is obtained by matching from a document template library according to the category of the candidate commodity word and/or the current period; and/or inputting at least one of the category and the current period of the candidate commodity word and the candidate commodity word into a pre-trained document generation model for processing, so as to obtain a target recommended document output by the document generation model.

In the embodiment, the generation of the recommended text based on different schemes is realized, the fatigue feeling of the user caused by the repeated display of the same text style is avoided, the display of the recommended search word in the text form is realized, the interaction feeling with the user can be enhanced, and the search enthusiasm of the user is improved.

In one implementation, the search page presents an avatar; the displaying the target recommended document in a search page comprises the following steps:

in the case that the target recommended text includes at least two, one of the target recommended text is displayed near the avatar in a manner simulating the speaking of the avatar, and after a preset period of time, the target recommended text is displayed in the search page in a bullet screen form, and the other target recommended text is continuously displayed near the avatar.

In this embodiment, the target recommended text is displayed by combining the virtual image speaking simulation mode and the barrage mode, so as to improve the interaction feeling with the user.

In one implementation, the presenting the target recommendation document in a search page includes:

under the condition that the target recommended texts comprise at least two target recommended texts, determining the display sequence of each target recommended text according to the category to which the candidate commodity word in the target recommended text belongs, and displaying the target recommended texts in the search page according to the display sequence; the category to which the candidate commodity word in the target recommended documents belongs is used for enabling at least two target recommended documents containing candidate commodity words belonging to the same category to be displayed discontinuously.

In this embodiment, at least two target recommended documents including candidate commodity words belonging to the same category are discontinuously displayed, so that fatigue caused by continuous occurrence of search words of the same category by a user is avoided, and diversity of recommendation is improved.

According to a second aspect of one or more embodiments of the present specification, there is provided an electronic device comprising:

a processor;

a memory for storing processor-executable instructions;

wherein the processor implements the method of any of the first or second aspects by executing the executable instructions.

According to a third aspect of one or more embodiments of the present description, a computer-readable storage medium is presented, having stored thereon computer instructions which, when executed by a processor, implement the steps of the method according to any of the first or second aspects.

Drawings

Fig. 1 is a schematic diagram of a search recommendation system according to an exemplary embodiment.

Fig. 2A is a flowchart of a search recommendation method according to an exemplary embodiment.

Fig. 2B is a flow chart illustrating another search recommendation method according to an exemplary embodiment.

Fig. 3 is a schematic diagram of a display page provided by an exemplary embodiment.

Fig. 4 is another flow chart of a search recommendation method according to an exemplary embodiment.

Fig. 5 is a schematic diagram of a search page provided by an exemplary embodiment.

FIG. 6 is a schematic diagram of a results presentation page provided by an exemplary embodiment.

Fig. 7 is a schematic flow chart of a search recommendation method according to an exemplary embodiment.

Fig. 8 is a schematic diagram of an apparatus according to an exemplary embodiment.

Detailed Description

Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with one or more embodiments of the present specification. Rather, they are merely examples of apparatus and methods consistent with aspects of one or more embodiments of the present description as detailed in the accompanying claims.

It should be noted that: in other embodiments, the steps of the corresponding method are not necessarily performed in the order shown and described in this specification. In some other embodiments, the method may include more or fewer steps than described in this specification. Furthermore, individual steps described in this specification, in other embodiments, may be described as being split into multiple steps; while various steps described in this specification may be combined into a single step in other embodiments.

Aiming at the problems in the related art, the embodiment of the specification provides a search recommendation method, which can respond to a search page display instruction, carry out data recall according to historical search data, acquire a plurality of candidate search words, realize recall of the search words meeting the search requirement and be beneficial to improving recommendation accuracy; according to the exposure click data of the historical recommended file, a first candidate file template is determined from the historical recommended file, so that file style forms which accord with user preference and/or popular preferences can be determined; generating at least one target recommended document according to the candidate search words and the first candidate document template; finally, displaying the target recommended text in the search page to realize the recommendation of search words in a text style form which accords with the preference of the user and/or popular among the public, thereby enhancing the interaction feeling with the user and improving the search enthusiasm of the user; and the electronic equipment can perform data search under the condition that the target recommended file is triggered, a user does not need to key in search words, operation steps of the user are reduced, and the search efficiency is improved.

Referring to fig. 1, fig. 1 is a search recommendation system provided in an embodiment of the present disclosure, where the search recommendation system includes a service end 100 and at least one client end 200, and the client end 200 may access the service end 100 through a network to use services provided by the service end 100, including but not limited to a commodity distribution service, a commodity purchase service, a reading service, an audio/video playing service, or a search service.

The server 100 may be a program installed in a background device to provide a service to a user. As shown in fig. 1, the background device may be a server, which may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDNs (Content Delivery Network, content distribution networks), and basic cloud computing services such as big data and artificial intelligence platforms.

The client 200 may be a program installed in a user device to provide services to a user, the client 200 including, but not limited to, an application APP, a Web page, an applet, a plug-in or component, and the like. As shown in fig. 1, the user devices include, but are not limited to, smartphones, personal digital assistants, tablet computers, personal computers, notebook computers, virtual reality terminal devices, augmented reality terminal devices, and the like.

The search recommendation method provided in the embodiment of the present disclosure may be executed by any one of the server 100 and the client 200, which is not limited in this embodiment. Taking the example that the server 100 executes the search recommendation method, the client 200 may send a search page display instruction to the server 100 in response to a display triggering operation of a user on a search page, the server 100 may execute the search recommendation method provided in the embodiment of the present disclosure in response to the search page display instruction received from the client 200, generate at least one target recommendation document, and then may send the search page displaying the target recommendation document to the client 200, so that the client 200 may display the search page, and further in a case that any one of the target recommendation documents in the search page is triggered, the client 200 may send a triggering instruction of the target recommendation document to the server 100, so that the server 100 may perform data search in response to the target recommendation document being triggered, and feedback the search result to the client 200.

Referring to fig. 2A, fig. 2A is a flowchart illustrating a search recommendation method according to an embodiment of the present disclosure. The method may be performed by an electronic device, which may be installed with a server or a client as described in the embodiment of fig. 1; the method comprises the following steps:

in S101, in response to the search page display instruction, a number of candidate search terms are recalled.

In S102, in a case where the plurality of candidate search terms includes a candidate commodity term, all target commodity terms corresponding to the commodity indicated by the candidate commodity term are determined.

In S103, if the user can be provided with the product named by the target product word within the distribution range of the user, a target recommended document is generated according to the candidate product word.

In S104, the target recommendation document is presented in a search page to perform a merchandise search if the target recommendation document is triggered.

Referring to fig. 2B, fig. 2B is a flowchart illustrating another search recommendation method according to an embodiment of the present disclosure. The method may be performed by an electronic device, which may be installed with a server or a client as described in the embodiment of fig. 1; the method comprises the following steps:

in S201, in response to the search page display instruction, data recall is performed according to the historical search data, and a plurality of candidate search terms are acquired.

In S202, a first candidate document template is determined from the history recommended documents according to the exposure click data of the history recommended documents.

In S203, at least one target recommended document is generated according to the plurality of candidate search terms and the first candidate document template.

In S204, the target recommendation document is presented in the search page for data searching if the target recommendation document is triggered.

In the embodiment, data recall is performed according to the historical search data, so that a plurality of candidate search words meeting the search requirement can be recalled, and the recommendation accuracy is improved; according to the exposure click data of the historical recommended file, a first candidate file template is determined from the historical recommended file, so that file style forms which accord with user preference and/or popular preferences can be determined; generating at least one target recommended document according to the candidate search words and the first candidate document template, and displaying the target recommended document in the search page to realize recommending the search words in a document style form which accords with the preference of the user and/or popular preferences of the user, so that the interaction feeling with the user is enhanced, and the search enthusiasm of the user is improved; and the electronic equipment can perform data search under the condition that the target recommended file is triggered, a user does not need to key in search words, operation steps of the user are reduced, and the search efficiency is improved.

It may be understood that the data recall process of step S201 and the first candidate document template determination process of step S202 may be performed in parallel, or may be performed sequentially, which is not limited in this embodiment.

In some embodiments, referring to fig. 3, a search bar may be displayed in a display page, and further, in response to a triggering operation (such as clicking, long pressing, or sliding, but not limited to) of the search bar by a user, a search page display instruction is generated. And the electronic equipment responds to the search page display instruction, executes the search recommendation method provided by the embodiment of the specification, and for example, carries out data recall according to historical search data to acquire a plurality of candidate search words.

In some possible embodiments, upon data recall, the electronic device may determine the search term and at least one reference information for the search term based on historical search data for other users in the vicinity of the user's location over a recent period of time (e.g., within a week or within 10 days, etc.); illustratively, the reference information includes, but is not limited to, at least one of: search period, weather information, holiday information, location information, etc., corresponding to the search term. And then determining the searching trend of the searching words according to at least one kind of reference information of the searching words, and determining the searching words with the upward searching trend as candidate searching words. The embodiment realizes that the search word which is liked or concerned by the public recently is determined as the candidate search word by combining the real-time information, and is favorable for recalling the search word which meets the search requirement of the user.

In another possible implementation manner, the server divides the user groups according to the characteristics of the users to which the different clients belong, so that targeted services are provided for different user groups, and the use experience of the users is improved. For example, preferences of different users belonging to the same user group are the same or similar, and the preference information of the users may be determined based on at least one parameter of the user's historical purchase behavior, historical search behavior, and historical browsing behavior; in other words, the same group of users indicates that one or more parameters of search behavior, purchase behavior, or browsing behavior of the users in the group are similar. After obtaining one or more parameters such as search behavior, purchase behavior or browsing behavior of the plurality of users, the parameters of the plurality of users may be clustered by using a clustering algorithm, so as to determine a user group to which the user belongs.

When the data recall is carried out, the electronic equipment can determine candidate search words according to at least one of historical behavior data of the user and historical behavior data of other users belonging to the same user group with the user; wherein the historical behavioral data includes, but is not limited to, at least one of: historical purchase behavior data, historical search behavior data, and historical browsing behavior data. The embodiment realizes recall of the search word conforming to the preference of the user and can conform to the search requirement of the user.

In one example, if the historical purchase behavior data of the user includes milk tea, cake, milk, etc., the milk tea, cake, and milk may be recalled as candidate search terms for the current search recommendation process.

In yet another possible implementation, the two recall modes can be combined, so that not only the search word which is loved or focused by the public in the near term, but also the search word which accords with the preference of the user can be recalled, and the recall result is more complete.

It will be appreciated that recalled search terms differ in different search scenarios.

For example, in a merchandise purchase scenario, the recalled candidate search term may be at least one of a merchandise term and a store term. Such as, but not limited to, in order scenes, commodity words such as milky tea, coffee, crayfish, roast, or hot dry noodles, etc.; the store words may be, but are not limited to, a mama cuisine, AA milk tea, or BB cantonese, etc. Also, in shopping scenarios, commodity words such as sweater, down jacket, dress, shirt, hair accessory, mobile phone or necklace, etc., but are not limited thereto; the store name may be, but is not limited to, a CC clothing house, DD clothing, EE jewelry store, or GG digital store, among others.

For example, and as in a reading scenario, the recalled candidate search term may be at least one of a novel name, an author, a hot news word, and a journal name. By way of example, also such as in an audiovisual scene, the recalled candidate search term may be at least one of a video title, a video genre, an opera title, a song title, a singer, a video creator name, and a presenter name.

In some embodiments, after recall of the number of candidate search words, to improve the recommendation of the candidate search words, the electronic device may perform quality detection on the recalled number of candidate search words and then filter out candidate search words that do not meet the preset quality condition. The quality condition may be specifically set according to the actual application scenario, which is not limited in this embodiment.

For example, the electronic device may classify each candidate search term using a pre-stored word stock and/or a pre-trained search term classification model, to determine a category of each candidate search term; and then deleting candidate search words which do not belong to the preset category. The word stock comprises a plurality of search words and categories thereof. The search word classification model is used for classifying and detecting the input search words to determine the categories of the input search words. It can be understood that the preset categories can be specifically set according to actual application scenes; for example, in a commodity purchase scene, the preset category includes at least one of a commodity and a store; also for example, in a reading scenario, the preset category includes at least one of a title and an author; but is not limited thereto.

In one example, for any one candidate search term, the electronic device may search the lexicon for a search term that matches it, and then determine the category of the matching search term as the category of the candidate search term.

In another example, for any one candidate search term, the electronic device may use a pre-trained search term classification model to classify the candidate search term, so as to obtain a category output by the search term classification model. The search word classification model can be obtained through training in a supervised training mode based on a search word sample and a category label of the search word sample. It can be appreciated that, in this embodiment, the specific model structure of the search term classification model is not limited, and may be specifically set according to an actual application scenario, for example, may be a deep learning model, so as to implement entity recognition.

In yet another example, for any one candidate search word, the electronic device searches the word stock for a search word that matches the candidate search word, and if the candidate search word has a matching search word in the word stock, determines the category of the matching search word as the category of the candidate search word; the word stock comprises a plurality of search words and categories thereof; and if the candidate search words do not have matched search words in the word stock, classifying the candidate search words by using a pre-trained search word classification model to obtain the categories output by the search word classification model. According to the embodiment, the category of the candidate search word is determined by combining the word bank and the search word classification model, so that the candidate search word which does not belong to the preset category is deleted, the quality of the candidate search word is guaranteed, and the subsequent recommendation efficiency is improved.

In some embodiments, after recalling a plurality of candidate search words, for some scenes with strong spatial features, the electronic device may further perform supply judgment on services indicated by each candidate search word, and determine whether the services indicated by the candidate search words can be provided for the user within the distribution range of the user; wherein, the distribution range of the user is determined according to the position information of the user; and then the candidate search words which can not provide service for the user can be deleted, so that invalid recommendation situations which can not meet the requirements of the user are avoided.

For example, in a take-out delivery scenario, the electronic device may perform a supply judgment on the goods indicated by each candidate search term, determine whether the goods indicated by the candidate search term may be provided in the delivery range of the user, and further delete the candidate search term that cannot provide the relevant goods in the delivery range of the user. In one example, if the candidate search term includes "coffee", and the supply judgment determines that coffee cannot be provided to the user within the distribution range of the user, the candidate search term of "coffee" can be filtered out, so that invalid recommendation condition that cannot meet the requirement of the user is avoided.

Wherein the search term includes at least one of a merchandise term and a store name. The candidate search term includes at least one of a candidate merchandise term and a candidate store name.

In one possible implementation manner, if the candidate search term is a candidate store name, the electronic device may determine whether the distribution range of the user is satisfied according to the location information of the store indicated by the candidate store name and the location information of the user; determining a distance between the store indicated by the candidate store name and the user according to the position information of the store and the position information of the user, and if the distance is smaller than the distribution distance indicated by the distribution range of the user, indicating that the service of the store can be provided for the user in the distribution range of the user; otherwise, it is determined that there is no offer within the delivery range of the user.

In one possible implementation manner, if the candidate word is a candidate commodity word, the electronic device may search for a store containing the commodity indicated by the candidate commodity word, for example, the electronic device may pre-store a correspondence between the commodity word and a store name, and may further search for a store containing the commodity indicated by the candidate commodity word from the pre-stored correspondence between the commodity word and the store name according to the candidate commodity word; and then determining whether the distribution range of the user is satisfied or not according to the position information of the store and the position information of the user.

Further, considering that the same commodity has a plurality of different calls or aliases, the recalled candidate commodity word in the embodiment of the present disclosure may be only one of the calls of the commodity, so as to avoid that the same commodity is different in call, and thus, in order to avoid that a supply relationship is misjudged, for example, a certain commodity is supplied in the distribution range of the user, only the call and the candidate commodity word are screened out, and thus, the judgment of the supply relationship is misjudged. To solve this problem, the present specification examples exemplify the following two possible implementations.

In one possible implementation, the electronic device may have a pre-stored trade name table that includes different calls for the same trade. After the candidate commodity words are obtained, the electronic equipment can obtain all target commodity words corresponding to the commodity indicated by the candidate commodity words from a prestored commodity name table according to the candidate commodity words; and searching a store containing the commodity indicated by the target commodity word from the corresponding relation between the prestored commodity word and the store name according to the target commodity word. In this embodiment, it is beneficial to avoid or reduce the occurrence of supply judgment errors caused by different calling methods of the same commodity, and improve the accuracy of supply judgment.

In another possible embodiment, a neural network model may be pre-trained for converting different calls for the same commodity into the same token vector. And, the electronic device may pre-store the correspondence between different calls (i.e., different commodity names) and the characterization vector of the same commodity. After obtaining the candidate commodity words, the electronic equipment can convert the candidate commodity words into target characterization vectors according to a pre-trained neural network model; then, according to the target characterization vector, determining all target commodity words corresponding to the target vector from the corresponding relation between the pre-stored characterization vector and commodity names; and finally searching the store containing the commodity indicated by the target commodity word from the corresponding relation between the prestored commodity word and the store name according to the target commodity word. In this embodiment, it is beneficial to avoid or reduce the occurrence of supply judgment errors caused by different calling methods of the same commodity, and improve the accuracy of supply judgment.

Illustratively, the neural network includes at least an embedded layer and an encoder; the embedding layer is used for converting the candidate commodity words to obtain an embedding vector; the encoder is used for mapping the embedded vector from a character vector space to a numerical vector space to obtain the characterization vector.

The neural network model can be obtained by comparison learning and characterization learning according to commodity samples with a plurality of different calls. Contrast learning (Contrastive Learning) belongs to one of the self-supervised learning, and the contrast learning learns the feature representation of the sample by comparing the data with positive and negative examples, respectively, in the feature space. Contrast learning focuses on learning common features between instances of the same class, distinguishing differences between instances of non-same class. Compared with the generation learning (Generative Learning), the comparison learning does not need to pay attention to complicated details on the examples, only needs to learn the distinction of data on the feature space of the abstract semantic level, so that the model and the optimization thereof are simpler, and the generalization capability is stronger. Token learning is a collection of techniques to learn a feature, converting raw data into a form that can be effectively exploited by machine learning. The method avoids the trouble of manually extracting the features, allows a computer to learn how to extract the features while learning the features: study how to learn.

In the training process, a plurality of binary group samples are obtained, wherein one part of binary group samples comprise two positive samples, and the other part of binary group samples comprise one positive sample and one negative sample; the two positive samples comprise different commodity names of the same commodity, and the negative sample and the positive sample belong to commodity names of different commodities; inputting the two-tuple samples into a preset neural network with two branches, and processing one of the two-tuple samples by each branch to obtain two characterization vectors; wherein the weights of the two branches are shared; according to the similarity between the characterization vectors respectively corresponding to the two positive samples and/or the difference degree between the characterization vectors of the positive samples and the characterization vectors of the negative samples, adjusting parameters of the preset neural network to obtain a trained neural network; wherein the trained neural network has at least one of the branches.

The optimization targets of the neural network model include: minimizing the distance between the characterization vectors respectively corresponding to different calls belonging to the same commodity sample and/or maximizing the distance between the characterization vectors respectively corresponding to at least two calls belonging to different commodity samples. In other words, in the training process of the neural network model, the neural network model learns a function F based on the commodity samples with a plurality of different names, and the function F can encode the input data into a characterization vector, so that the characterization vectors respectively corresponding to the different names belonging to the same commodity sample are as similar as possible, and the characterization vectors respectively corresponding to at least two trade names belonging to different commodity samples are as different as possible, thereby improving the accuracy of the model.

In some embodiments, after recalling a number of candidate search words, the electronic device may screen out the target search word from the number of candidate search words based on historical exposure click data for the different candidate search words. For any user, the search word displayed on the search page of the client of the user (i.e. the exposure condition of the search word) and the feedback condition of the user for the displayed search word (such as the click condition of the search word) can be recorded in the user behavior log corresponding to the user; in order to improve the recommendation accuracy, the historical exposure click data of the different candidate search words may be obtained from the user behavior log of the client and/or the user behavior log of other users belonging to the same user group as the user.

For example, the electronic device may determine, from the number of candidate search words, candidate search words that have been exposed and clicked during the history presentation, and candidate search words that have been exposed and not clicked during the history presentation, based on the history exposure click data of the different candidate search words; wherein the exposed and clicked candidate search terms reflect the user's preferences and the exposed and non-clicked candidate search terms indicate that the user may not be interested in them, the target search terms that are finalized may include at least a portion of the exposed and clicked candidate search terms during the history presentation, and a portion of the non-exposed candidate search terms, and the target search terms do not include the exposed and non-clicked candidate search terms during the history presentation. Screening candidate search words based on user preferences is achieved, and candidate search words which can meet user search requirements are determined; and the target search term contains unexposed candidate search terms, which helps explore and mine user interest.

In some embodiments, the electronic device may generate a target recommended document according to the candidate search term, so as to implement recommending the search term in a document form, where the target recommended document may be a personified document (for example, a document in a dialogue form), so as to enhance interaction feeling with the user and improve search enthusiasm of the user.

In a first possible implementation, the electronic device pre-stores the document templates under different time periods and/or different classes; for example, refer to table 1, which shows a document template in a take-away scenario; wherein "$keyword" in table 1 is used to indicate the search term to be inserted. Exemplary, such as in a take-away scenario, such as 6:00-10:00 breakfast time, 11:00-13:00 lunch time, 15:00-16:00 afternoon tea time, 17:00-20:00 dinner time, and 22:00-24:00 overnight time.

After recalling a plurality of candidate search words, the electronic equipment can obtain a second candidate document template from the document template library according to at least one of the category to which the candidate search word belongs and the current period of time for any one candidate search word; and then generating a target recommended document according to the candidate search word and the matched second candidate document template. The embodiment realizes that the search word is recommended in a preset text form, so that the interaction feeling with the user can be enhanced.

TABLE 1

In a second possible implementation, a document generation model may be trained in advance according to actual needs; for example, the method can be used for performing supervised training based on a plurality of samples and labels thereof to obtain a document generation model; any sample includes at least one of a category to which the reference search term belongs and a reference period, and the reference search term, and the sample is labeled with a recommended document including the reference search term. The trained document generation model is used for generating a recommended document according to at least one of the category and the reference period to which the output search word belongs and the reference search word.

In the practical application process, after recalling a plurality of candidate search words, for any one candidate search word, the electronic device can input at least one of the category and the current period to which the candidate search word belongs and the candidate search word into a pre-trained document generation model for processing, so as to obtain a target recommended document output by the document generation model. The embodiment realizes that the search word is recommended in a text form, so that the interaction feeling with the user can be enhanced.

In a third possible implementation manner, the electronic device may determine a first candidate document template from the historical recommended document according to the exposure click data of the historical recommended document, and further generate at least one target recommended document according to the plurality of candidate search terms and the first candidate document template.

For example, for any user, the recommended document (i.e., the exposure condition of the recommended document) displayed on the search page of the client of the user and the feedback condition of the user for the displayed recommended document (such as the click condition of the recommended document) may be recorded in the user behavior log corresponding to the user; in order to improve the recommendation accuracy, the historical recommendation file and the exposure click data of the historical recommendation file may be obtained from at least one of the user behavior log of the client, the user behavior log of other users belonging to the same user group with the user, and the user behavior log of other users.

Illustratively, the historical recommendation document includes at least one of: (1) The recommended text is displayed in a search page of a client of a user; (2) The recommended text is displayed on a search page of a client of other users belonging to the same user group as the user; wherein preferences of different users belonging to the same user group are the same or similar; (3) Recommended documents displayed on the search pages of clients of other users. Illustratively, the first candidate document template is generated from historical recommended documents that have been exposed and clicked on during the history presentation. According to the embodiment, the text style form which accords with the preference of the user and/or popular among the public is determined according to the exposure click condition of the historical recommended text, so that search words can be recommended in the text style form which accords with the preference of the user and/or popular among the public, the interaction feeling with the user is enhanced, and the search enthusiasm of the user is improved.

In one example, where the history recommended notes that have been exposed and clicked during the history display include "late night eat barbecue, too cool cheer" and "late night delicious baby recommended fry Niu He", etc., the first candidate pattern template generated may be "late night eat $keyword, too cool cheer" and "late night delicious baby recommended $keyword". Wherein "$keyword" is used to indicate the search term to be inserted. Assuming that the candidate search terms include crayfish and roast meat, the target recommended text generated may be "eat crayfish overnight, too cool cheer" and "delicious kibble late at night recommended roast meat".

In an exemplary embodiment, referring to fig. 4, for any one user, a recommended document displayed on a search page of a client of the user and a feedback condition of the user for the displayed recommended document may be recorded in a user behavior log corresponding to the user. In other words, the user behavior log records the search word contained in the recommended document and the exposure click data thereof, and the recommended document and the exposure click data thereof.

After recalling a plurality of candidate search words, the electronic equipment can firstly acquire historical exposure click data of different candidate search words from a user behavior log of the client, and then screen out target search words from the plurality of candidate search words according to the historical exposure click data of the different candidate search words; for example, the target search term may include a candidate search term that has been at least partially exposed and clicked during the history presentation, and a candidate search term that has not been partially exposed, and the target search term does not include a candidate search term that has been exposed and not clicked during the history presentation; the method and the device realize screening of candidate search words based on user preferences, and determine the candidate search words capable of meeting the search requirements of users.

Further, the electronic device may obtain exposure click data of the history recommended document from the user behavior log of the client, and then determine a first candidate document template from the history recommended document according to the exposure click data of the history recommended document; illustratively, the first candidate document template is generated according to the history recommended document which is exposed and clicked in the history display process, so that the document style form preferred by the user is obtained.

In the case that there are a plurality of target search words, the plurality of target search words may be divided into different portions, and then the documents may be generated for the target search words of each portion according to the above-described different generation manners of the target recommended documents. In one example, multiple target search terms may be divided into 3 portions, for example; for the target search term of the first portion, the electronic device may generate at least one target recommended document from the target search term and the first candidate document template. And for the target search word of the second part, the electronic equipment matches from the document template library to obtain a second candidate document template according to at least one of the category to which the target search word belongs and the current period, and then generates a target recommended document according to the target search word and the matched second candidate document template. And for the target search word of the third part, the electronic equipment can input at least one of the category and the current period of the target search word and the target search word into a pre-trained document generation model for processing, so as to obtain a target recommended document output by the document generation model. According to the embodiment, the recommended text is generated based on different schemes, the fatigue of the user caused by repeated display of the same text style is avoided, the recommended search words are displayed in the text form, the interaction feeling with the user can be enhanced, and the search enthusiasm of the user is improved.

After obtaining a plurality of target recommended documents, the electronic equipment can sort the target recommended documents according to preference information of the user so as to display the target recommended documents in the search page according to the sorting result; the preference information of the user can be determined according to the exposure click data of the history recommended file obtained from the user behavior log of the client. The user can trigger the target recommended file according to the actual requirement, and the electronic equipment can search data under the condition that the target recommended file is triggered, so that the user does not need to key in search words, the operation steps of the user are reduced, and the search efficiency is improved. For example, the target recommended document displayed in the search page and the feedback of the user on the target recommended document (such as the click condition of the user on the target recommended document) may be recorded in the user behavior log corresponding to the user, so as to be used as the reference data of the processes of subsequent candidate search word screening, target recommended document generation, target document sorting and the like.

In some embodiments, after generating the target recommended documents, the electronic device may further detect document quality of each target recommended document to obtain target recommended documents meeting the document quality condition, and further may display the target recommended documents meeting the document quality condition in the search page.

In one possible implementation, an emotion classification model may be pre-trained for emotion analysis of the input text to determine an emotion classification of the input text, including positive emotion, neutral emotion, or negative emotion. Supervised training can be performed based on several text samples and their emotion labels (forward emotion, neutral emotion, or negative emotion) to obtain emotion classification models. It can be appreciated that the specific model result of the language model in this embodiment of the present disclosure is not limited, and may be specifically set according to an actual application scenario, for example, the emotion classification model may be a bert (Bidirectional Encoder Representation from Transformers, bi-directional coding from convertors) model or other deep learning model.

In the actual application process, the electronic equipment can process the target recommended file by utilizing a pre-trained emotion classification model to obtain emotion classification information of the target recommended file. For example, target recommendation documents whose emotion classification information is "negative emotion" may be filtered out. In one example, if the target recommended document is "milk tea is not drunk well", and the emotion classification information of the target recommended document is determined to be "negative emotion" based on the emotion classification model, the target recommended document of "milk tea is not drunk well" may be deleted.

In another possible implementation, a language model may be pre-trained that measures the rationality of a sentence (judging natural language context-dependent characteristics, sentence fluency). The language model can be obtained by performing supervised training based on a plurality of text samples and fluency labels thereof. It will be appreciated that the specific model result of the language model in this embodiment of the present disclosure is not limited, and may be specifically set according to an actual application scenario, for example, the language model may be a bert (Bidirectional Encoder Representation from Transformers, bi-directional coding from converters) model or other deep learning model.

In the practical application process, the electronic device can detect the text fluency of the target recommended document by using a pre-trained language model, and obtain the text fluency of the target recommended document. For example, the target recommended text whose text fluency does not meet the preset fluency requirement can be filtered out. In one example, if the target recommended document is "the crisp shrimp cake is popular", and the text fluency of the target recommended document is determined to be not in accordance with the preset fluency requirement based on the language model, the target recommended document of "the crisp shrimp cake is popular" may be deleted.

In yet another possible implementation manner, a relevance evaluation model may be trained in advance, where the relevance evaluation model is used to evaluate whether the target recommended document includes a search term, and/or whether the target recommended document accords with the current period, and output a relevance evaluation result; illustratively, the correlation evaluation result includes "yes" or "no".

For example, several training samples may be obtained, any training sample including a reference document and a reference period thereof, any training sample corresponding to a positive label or a negative label, the positive label indicating that the reference document contains a search term and/or that the reference document corresponds to the reference period, such as a positive label of "yes" or "1"; negative labels indicate that the reference document does not contain a search term or that the reference document does not conform to a reference period, such as negative labels of "no" or "0"; and further, the supervised training can be performed based on a plurality of training samples and labels (positive labels or negative labels) thereof, so as to obtain a correlation evaluation model. Of course, the positive label and the negative label may be other representations, which are not limited in this embodiment.

In the actual application process, the electronic device may process the target recommended document and the current period by using a pre-trained relevance evaluation model, to obtain a relevance evaluation result of the target recommended document, where the relevance evaluation result is used to indicate whether the target recommended document includes a search term, and/or whether the target recommended document accords with the current period. In one example, if the correlation evaluation result is "no", it indicates that the target recommended document does not include a search term, and/or the target recommended document does not conform to the current period; for example, if the correlation evaluation result is "yes", it indicates that the target recommended document contains a search term, and the target recommended document accords with the current period. Of course, the correlation evaluation result may be other representation manners, which are not limited in this embodiment.

In an exemplary embodiment, emotion may be classified as forward emotion or neutral emotion, text fluency of a target recommended document meets a preset fluency requirement, and a correlation evaluation result of the target recommended document meets a preset correlation requirement (the preset correlation requirement indicates that the target recommended document contains a search term, and the target recommended document meets a current scene), which is displayed in the search page, thereby facilitating improvement of recommendation accuracy.

In some embodiments, after the target recommended text is generated, the target recommended text can be displayed in the search page, so that the search word is recommended to the user in an interactive text form, the interaction feeling with the user is enhanced, and the search enthusiasm of the user is improved; and the electronic equipment can perform data search under the condition that the target recommended file is triggered (for example, the target recommended file is clicked by the user so that the target recommended file is triggered), search words are not required to be typed by the user, the operation steps of the user are reduced, and the search efficiency is improved.

For example, in the case that the target recommended documents include at least two target recommended documents, the electronic device may determine, according to preference information of a user and a category to which a target search word in each target recommended document belongs, a display order of each target recommended document, so as to display the target recommended documents in the search page according to the display order; the category to which the target search word belongs is used for enabling at least two target recommended documents containing the target search words belonging to the same category to be discontinuously displayed, so that fatigue caused by continuous appearance of the search words of the same category by a user is avoided; the preference information is determined according to at least one of a user's historical purchase behavior, a historical search behavior, and a historical browsing behavior.

For example, referring to fig. 5, the electronic device may display the target recommended document in the search page in a bullet screen. In one example, the target recommended documents may be scrolled and displayed in the search page in a bullet screen according to the display order of the target recommended documents determined in the above process.

For example, referring to fig. 5, the search page displays an avatar, and the electronic device may display the target recommended text near the avatar in a manner simulating speaking of the avatar, thereby facilitating improvement of interaction feeling with the user and improving enthusiasm of the user for triggering the recommended text.

Further, in case that the target recommended document includes at least two, after one of the target recommended documents is presented in the vicinity of the avatar for a preset period of time in a manner simulating the talking of the avatar, the target recommended document may be displayed in the search page in a bullet screen form and another target recommended document is continuously presented in the vicinity of the avatar. Wherein the target recommended documents displayed in the vicinity of the avatar in a manner simulating the avatar speaking may be displayed in the display order determined in the above-described process. The specific value of the preset duration can be specifically set according to the actual application scenario, and the embodiment does not limit the specific value.

For example, referring to fig. 6, the electronic device may obtain, in response to a trigger instruction for the target recommended document, search results related to search terms included in the target recommended document; displaying the search result in a result display page; the result display page further comprises a search column, and the search column displays search words contained in the target recommended text so that a user can clearly determine a search object in the current search process. For example, the user clicks on the target recommended text "eating the bar in rainy days" in fig. 5, and the "bar" in the target recommended text is shown in the search bar in fig. 6.

For example, the target recommended document shown in the search page and the feedback of the target recommended document by the user (such as the trigger on the target recommended document) may be recorded in the user behavior log corresponding to the user. In other words, the user behavior log records the search word and the exposure click data thereof, the recommended document and the exposure click data thereof contained in the recommended document so as to be used as reference data for the subsequent processes of candidate search word screening, target recommended document generation, target document ordering and the like.

The various technical features of the above embodiments may be arbitrarily combined as long as there is no conflict or contradiction between the features, but are not described in detail, and therefore, the arbitrary combination of the various technical features of the above embodiments is also within the scope of the disclosure of the present specification.

Accordingly, referring to fig. 7, fig. 7 is a flowchart of another search recommendation method provided in the embodiment of the present disclosure, where the method may be performed by a client (or an electronic device with the client installed), and the method includes:

in S301, at least one target recommended document is acquired in response to a search page display instruction; the target recommendation document includes search terms.

In S302, displaying the target recommended document in the search page in a bullet screen form; and/or the search page is provided with an avatar, and the target recommended text is provided near the avatar in a mode of simulating the speaking of the avatar.

The embodiment realizes that the search word is recommended in a text form, so that the interaction feeling with the user can be enhanced, and the search enthusiasm of the user is improved.

In some embodiments, the presenting the target recommendation document in proximity to the avatar in a manner that simulates speaking of the avatar includes:

in the case where the target recommended text includes at least two, after one of the target recommended text is presented in the vicinity of the avatar for a preset period of time in a manner simulating the avatar speaking, the target recommended text is presented in the search page in the form of a bullet screen, and the other target recommended text is presented in the vicinity of the avatar.

In some embodiments, further comprising: under the condition that the target recommended documents comprise at least two, determining the display sequence of each target recommended document according to preference information of a user and the category of the target search word in each target recommended document, so as to display the target recommended document in the search page according to the display sequence; the category to which the target search word belongs is used for enabling at least two target recommended documents containing the target search word belonging to the same category to be discontinuously displayed; the preference information is determined according to at least one of a user's historical purchase behavior, a historical search behavior, and a historical browsing behavior.

In some embodiments, the obtaining at least one target recommendation document includes: carrying out data recall according to the historical search data to obtain a plurality of candidate search words; determining a first candidate document template from the history recommended document according to exposure click data of the history recommended document; and generating at least one target recommended document according to the candidate search words and the first candidate document template.

In some embodiments, after the data recall is performed according to the historical search data, the method further comprises: for each candidate search word, carrying out supply judgment on the service indicated by the candidate search word, and determining whether the service indicated by the candidate search word can be provided for the user in the distribution range of the user; wherein the distribution range of the user is determined according to the position information of the user; candidate search terms that cannot serve the user are deleted.

In some embodiments, the search term includes a merchandise term and/or a store name; the step of determining whether the service indicated by the candidate search word can be provided for the user within the distribution range of the user by performing supply judgment on the service indicated by the candidate search word comprises the following steps: if the candidate search word is a candidate store name, determining whether the distribution range of the user is met or not according to the position information of the store indicated by the candidate store name and the position information of the user; if the candidate word is a candidate commodity word, searching a store containing the commodity indicated by the candidate commodity word, and determining whether the distribution range of the user is met according to the position information of the store and the position information of the user.

In some embodiments, the searching for a store containing the item indicated by the candidate item word includes: according to the candidate commodity words, acquiring all target commodity words corresponding to the commodity indicated by the candidate commodity words from a prestored commodity name list; different calls of the same commodity are prestored in the commodity name table; and searching a store containing the commodity indicated by the target commodity word from the corresponding relation between the prestored commodity word and the store name according to the target commodity word.

In some embodiments, the searching for a store containing the item indicated by the candidate item word includes: converting the candidate commodity words into target characterization vectors according to a pre-trained neural network model; the neural network model is used for converting different calls of the same commodity into the same characterization vector; determining all target commodity words corresponding to the target vector from the corresponding relation between the pre-stored characterization vector and commodity names according to the target characterization vector; and searching a store containing the commodity indicated by the target commodity word from the corresponding relation between the prestored commodity word and the store name according to the target commodity word. In some embodiments, the historical recommendation document includes at least one of: the recommended text is displayed in a search page of a client of a user; the recommended text is displayed on a search page of a client of other users belonging to the same user group as the user; wherein preferences of different users belonging to the same user group are the same or similar; recommended documents displayed on the search pages of clients of other users. Wherein the first candidate document template is generated from historical recommended documents that have been exposed and clicked during the historical display process.

In some embodiments, after the data recall is performed according to the historical search data, the method further comprises: screening target search words from the plurality of candidate search words according to historical exposure click data of different candidate search words; wherein the target search term includes candidate search terms that have been exposed and clicked at least in part during the history presentation process and does not include candidate search terms that have been exposed and not clicked during the history presentation process.

Generating at least one target recommended document according to the plurality of candidate search words and the first candidate document template, including: and generating at least one target recommended file according to the target search word and the first candidate file template. In some embodiments, further comprising: for any one candidate search word, matching from a document template library according to at least one of the category to which the candidate search word belongs and the current period to obtain a second candidate document template, wherein the document template library comprises document templates under different periods and/or different categories; and generating a target recommended document according to the candidate search word and the second candidate document template obtained by matching.

In some embodiments, further comprising: for any one candidate search word, inputting at least one of the category and the current period of the candidate search word and the candidate search word into a pre-trained document generation model for processing, and obtaining a target recommended document output by the document generation model; the document generation model is obtained based on a plurality of samples and labels thereof; any sample includes at least one of a category to which the reference search term belongs and a reference period, and the reference search term, and the sample is labeled with a recommended document including the reference search term.

In some embodiments, the target recommended documents shown in the search page satisfy document quality conditions. The method further comprises the steps of: after obtaining the target recommended documents, detecting document quality of each target recommended document to obtain target recommended documents meeting document quality conditions.

The detecting of the document quality of each target recommended document includes: processing the target recommended file by using a pre-trained emotion classification model to obtain emotion classification information of the target recommended file; the emotion classification model is used for carrying out emotion analysis on the input text to determine emotion classification of the input text, wherein the emotion classification comprises forward emotion, neutral emotion or negative emotion; and/or detecting the text fluency of the target recommended document by using a pre-trained language model to obtain the text fluency of the target recommended document; and/or processing the target recommended file and the current period by utilizing a pre-trained relevancy assessment model to acquire a relevancy assessment result of the target recommended file; the relevance evaluation model is used for evaluating whether the target recommended file contains search words and/or whether the target recommended file accords with the current time period.

In some embodiments, the performing data recall according to the historical search data to obtain a plurality of candidate search terms includes: determining search words and at least one type of reference information of the search words according to historical search data of other users in the recent period of time, wherein the other users are near the position of the user; determining a search trend of the search word according to at least one piece of reference information of the search word, and determining the search word with the upward search trend as a candidate search word; and/or determining candidate search terms according to at least one of historical behavior data of the user and historical behavior data of other users belonging to the same user group with the user; wherein the historical behavior data includes at least one determination of historical purchase behavior data, historical search behavior data, and historical browsing behavior data.

In some embodiments, after the data recall is performed according to the historical search data, the method further comprises: classifying each candidate search word by utilizing a pre-stored word stock and/or a pre-trained search word classification model, and determining the category of each candidate search word; the word stock comprises a plurality of search words and categories thereof; the search word classification model is used for carrying out classification detection on the input search words so as to determine the categories of the input search words; and deleting the candidate search words which do not belong to the preset category.

In some embodiments, further comprising: responding to a trigger instruction of the target recommended file, and acquiring search results related to search words contained in the target recommended file; displaying the search results in a result display page; the result display page further comprises a search column, and the search column displays search words contained in the target recommended text.

For the above embodiments, please refer to the description of the method embodiment of fig. 2A, and the description is omitted here.

Fig. 8 is a schematic block diagram of an apparatus according to an exemplary embodiment. Referring to fig. 8, at the hardware level, the device includes a processor 802, an internal bus 804, a network interface 806, a memory 808, and a non-volatile storage 810, although other hardware required by the service is also possible. One or more embodiments of the present description may be implemented in a software-based manner, such as by the processor 802 reading a corresponding computer program from the non-volatile memory 810 into the memory 808 and then running. Of course, in addition to software implementation, one or more embodiments of the present disclosure do not exclude other implementation manners, such as a logic device or a combination of software and hardware, etc., that is, the execution subject of the following processing flow is not limited to each logic unit, but may also be hardware or a logic device.

Accordingly, the embodiment of the present specification further provides an electronic device, including:

a processor;

a memory for storing processor-executable instructions;

wherein the processor implements the method of any of the above by executing the executable instructions.

Accordingly, the present disclosure also provides a computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of any of the methods described above.

Accordingly, the present disclosure also provides a computer program product which, when executed by a processor, implements the steps of any of the methods described above.

It should be noted that, the user information (including, but not limited to, user equipment information, user personal information, etc.) and the data (including, but not limited to, data for analysis, stored data, presented data, etc.) related to the present disclosure are information and data authorized by the user or sufficiently authorized by each party, and the collection, use and processing of the related data need to comply with the related laws and regulations and standards of the related country and region, and be provided with corresponding operation entries for the user to select authorization or rejection.

The system, apparatus, module or unit set forth in the above embodiments may be implemented in particular by a computer chip or entity, or by a product having a certain function. A typical implementation device is a computer, which may be in the form of a personal computer, laptop computer, cellular telephone, camera phone, smart phone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or a combination of any of these devices.

In a typical configuration, a computer includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.

The memory may include volatile memory in a computer-readable medium, random Access Memory (RAM) and/or nonvolatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of computer-readable media.

Computer readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of storage media for a computer include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, read only compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic disk storage, quantum memory, graphene-based storage or other magnetic storage devices, or any other non-transmission medium, which can be used to store information that can be accessed by the computing device. Computer-readable media, as defined herein, does not include transitory computer-readable media (transmission media), such as modulated data signals and carrier waves.

It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or apparatus that comprises the element.

The foregoing describes specific embodiments of the present disclosure. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

The terminology used in the one or more embodiments of the specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of the specification. As used in this specification, one or more embodiments and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used herein refers to and encompasses any or all possible combinations of one or more of the associated listed items.

It should be understood that although the terms first, second, third, etc. may be used in one or more embodiments of the present description to describe various information, these information should not be limited to these terms. These terms are only used to distinguish one type of information from another. For example, first information may also be referred to as second information, and similarly, second information may also be referred to as first information, without departing from the scope of one or more embodiments of the present description. The word "if" as used herein may be interpreted as "at … …" or "at … …" or "responsive to a determination", depending on the context.

The foregoing description of the preferred embodiment(s) is (are) merely intended to illustrate the embodiment(s) of the present invention, and it is not intended to limit the embodiment(s) of the present invention to the particular embodiment(s) described.

Claims

1. A search recommendation method, comprising:

2. The method of claim 1, wherein the determining all target commodity words corresponding to the commodity indicated by the candidate commodity word comprises:

3. The method of claim 2, wherein the determining that the results of the conversion using the neural network model are each a target commodity word of the target token vector comprises:

4. The method of claim 2, wherein the neural network model comprises an embedded layer and an encoder; the target characterization vector is obtained by converting the candidate commodity words into embedded vectors through the embedding layer and mapping the embedded vectors from a character vector space to a numerical vector space through the encoder.

5. The method of claim 2, wherein training the neural network model comprises:

6. The method of claim 1, wherein the determining all target commodity words corresponding to the commodity indicated by the candidate commodity word comprises:

7. The method according to any one of claims 1 to 6, wherein generating a target recommended document from the candidate commodity word further comprises, before if a commodity named by the target commodity word can be provided to the user within the delivery range of the user:

8. The method as recited in claim 1, further comprising:

9. An electronic device, comprising:

a processor;

a memory for storing processor-executable instructions;

wherein the processor is configured to implement the method of any one of claims 1 to 8 by executing the executable instructions.

10. A computer readable storage medium having stored thereon computer instructions which, when executed by a processor, implement the steps of the method of any of claims 1 to 8.