CN115827841A

CN115827841A - Searching method and device

Info

Publication number: CN115827841A
Application number: CN202211503431.0A
Authority: CN
Inventors: 刘相言; 杜荣
Original assignee: Shanghai Bilibili Technology Co Ltd
Current assignee: Shanghai Bilibili Technology Co Ltd
Priority date: 2022-11-28
Filing date: 2022-11-28
Publication date: 2023-03-21

Abstract

The embodiment of the application provides a searching method, which comprises the following steps: acquiring a first search word input by a target user; rewriting the first search word according to a preset rewriting word list to obtain N second search words, wherein the rewriting word list is obtained according to historical search words of a plurality of users in the same session, and N is a positive integer; and determining the associated commodity recommended to the target user according to the second search word, and returning the associated commodity. The searching method provided by the embodiment of the application can enable the commodities recommended at the bottom of the search to be related to the search terms of the user, the recommended commodities meet the requirements of the user, the interest of clicking or purchasing by the user is easily aroused, and the conversion rate is improved.

Description

Searching method and device

Technical Field

The present application relates to the field of internet technologies, and in particular, to a search method, an apparatus, a computer device, and a storage medium.

Background

The search bottom (feedblast) is a module below the search result page, and supports the flow when the user continues to slide downwards after finishing viewing the search result.

In the search result page, because the user has a relatively strong search intention, the recommended goods in the search bottom should have a certain correlation with the search term (query) of the user, so that the recommended goods can meet the requirements of the user, and further the conversion rate of the recommended goods is improved. However, at present, only users and commodities are considered in the recommendation logic of the search bottom in the search pages of most platforms, so that the commodities recommended at the search bottom are irrelevant to the search terms of the users in many cases, the recommended commodities do not meet the requirements of the users, the interest of the users in clicking or purchasing cannot be aroused, and the conversion rate is low.

Disclosure of Invention

The application aims to provide a searching method, a searching device, computer equipment and a storage medium, which are used for solving the technical problem that the conversion rate of bottom recommended commodities is low in the conventional searching.

One aspect of an embodiment of the present application provides a search method, including: acquiring a first search word input by a target user; rewriting the first search word according to a preset rewriting word list to obtain N second search words, wherein the rewriting word list is obtained according to historical search words of a plurality of users in the same session, and N is a positive integer; and determining the associated commodity recommended to the target user according to the second search word, and returning the associated commodity.

Optionally, the method further comprises: acquiring first historical search terms and second historical search terms of a plurality of users in the same session, wherein the first historical search terms are historical search terms of which the corresponding click rates meet a first preset condition or the corresponding conversion rates meet a second preset condition, and the second historical search terms are historical search terms of which the corresponding click rates do not meet the first preset condition and the corresponding conversion rates do not meet the second preset condition; forming a word pair set by combining each first historical search word and each second historical search word respectively, wherein the word pair set comprises a plurality of groups of word pairs; determining whether the word pairs in the word pair set meet a third preset condition, and saving the current word pairs to the rewritten word list under the condition that the current word pairs meet the third preset condition, wherein the third preset condition at least comprises that the co-occurrence frequency of the word pairs is greater than or equal to a frequency threshold value.

Optionally, the third preset condition further includes that the similarity of the word pair is greater than or equal to a similarity threshold.

Optionally, before the determining whether a word pair in the word pair set satisfies a third preset condition, the method further includes: obtaining semantic vectors of the first historical search word and the second historical search word in the word pair; determining similarity of the first historical search word and the second historical search word in the word pair according to the semantic vector.

Optionally, the determining, according to the second search term, the associated goods recommended to the target user includes: dividing the N second search terms into a first priority queue and a second priority queue, wherein the first priority queue and the second priority queue respectively comprise a plurality of second search terms; recalling the first commodity according to a recall submodel in the search model and the second search terms in the first priority queue; and determining the associated commodity according to the search model and the first commodity when the quantity of the first commodity is greater than or equal to a quantity threshold value or the current request concurrency number is greater than or equal to a concurrency threshold value.

Optionally, the method further comprises: under the condition that the number of the first commodities is smaller than the number threshold value and the request times are smaller than preset times, recalling second commodities by adopting the recall sub-model according to the second search terms in the second priority queue; determining the associated commodity according to the search model, the first commodity and the second commodity.

Optionally, the method further comprises: under the condition that the number of the second search words in the first priority queue is smaller than a first number, acquiring relevant words of the second search words in the first priority queue from a knowledge graph to perform candidate selection; or, when the number of the second search terms in the second priority queue is smaller than a second number, acquiring relevant terms of the second search terms in the second priority queue from the knowledge graph to perform candidate selection.

Optionally, the determining, according to the second search term, the associated goods recommended to the target user includes: acquiring a third commodity output by the first sequencing submodel in the search model by adopting a second sequencing submodel in the search model, wherein the third commodity is a commodity determined by the search model according to the second search word; grading the third commodity by adopting the second sequencing submodel to obtain a sequencing score of the third commodity; respectively obtaining the similarity between a second search word corresponding to the third commodity and the first search word, and determining the similarity score of the third commodity according to the similarity between the second search word and the first search word; determining a final score of the third commodity according to the ranking score and the similarity score; and taking the third commodity with the final score larger than a score threshold value as the associated commodity.

An aspect of an embodiment of the present application further provides a search apparatus, including: the acquisition module is used for acquiring a first search word input by a target user; the rewriting module is used for rewriting the first search word according to a preset rewriting word list to obtain N second search words, the rewriting word list is obtained according to historical search words of a plurality of users in the same session, and N is a positive integer; and the determining module is used for determining the associated commodity recommended to the target user according to the second search term and returning the associated commodity.

An aspect of the embodiments of the present application further provides a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-mentioned search method.

An aspect of the embodiments of the present application further provides a computer-readable storage medium, in which a computer program is stored, where the computer program is executable by at least one processor to cause the at least one processor to execute the steps of the above-mentioned search method.

The searching method, the searching device, the computer equipment and the storage medium provided by the embodiment of the application have the following advantages:

the method comprises the steps that a first search word input by a target user is obtained, the first search word is rewritten according to a preset rewriting word list, and N second search words are obtained, wherein the rewriting word list is obtained according to historical search words of a plurality of users in the same conversation; determining the associated commodities recommended to the target user according to the second search terms, and returning the associated commodities; because the rewriting vocabulary is obtained through the historical search words of a plurality of users in the same session, and the historical search words of the users in the same session have correlation, the original search words input by the users are rewritten through the rewriting vocabulary, and then the related goods are obtained according to the rewritten search words and recommended to the users, so that the goods recommended at the bottom of the search (fed blast) and the search words of the users have certain correlation, the recommended related goods meet the requirements of the users, the interest of clicking or purchasing by the users is easy to arouse, and the conversion rate is improved.

Drawings

Fig. 1 is a diagram schematically showing an environment architecture of a search method according to an embodiment of the present application;

FIG. 2 is a flow chart schematically illustrating a searching method according to a first embodiment of the present application

FIG. 3 is a diagram illustrating an application example of a search method according to an embodiment of the present application;

FIG. 4 is a flow chart of the addition step of FIG. 2;

FIG. 5 is a flow chart of the step of adding of FIG. 4;

FIG. 6 is a diagram illustrating an example of an application of a search method to find a rewritten word in an embodiment of the present application;

FIG. 7 is a flowchart of sub-steps of step S430 in FIG. 2;

FIG. 8 is a flow chart of the adding step of FIG. 7;

FIG. 9 is a flowchart of another substep of step S430 of FIG. 2;

FIG. 10 is a flowchart illustrating a searching method according to an embodiment of the present application;

fig. 11 is an exemplary diagram of a search result page obtained by the search method according to the embodiment of the present application;

fig. 12 is a block diagram schematically showing a search apparatus according to a second embodiment of the present application;

fig. 13 schematically shows a hardware architecture diagram of a computer device according to a third embodiment of the present application.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

It should be noted that the descriptions relating to "first", "second", etc. in the embodiments of the present application are only for descriptive purposes and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In addition, technical solutions between various embodiments may be combined with each other, but must be realized by a person skilled in the art, and when the technical solutions are contradictory or cannot be realized, such a combination should not be considered to exist, and is not within the protection scope of the present application.

In the description of the present application, it should be understood that the numerical references before the steps do not identify the order of performing the steps, but merely serve to facilitate the description of the present application and to distinguish each step, and therefore should not be construed as limiting the present application.

The following are explanations of terms referred to in the present application:

searching bottom feedblast: search results page the modules below the search results, such as the "recommend you" item list exposed below the slide to search results.

query: search terms searched by the user.

session: one-time session behavior of the user is identified, and the user can continuously span multiple channels, for example, recommendation and search are accessed successively, and the user can be regarded as one-time session.

Click rate (ctr): the ratio of the number of times a certain content on a website page is clicked to the number of times the certain content is displayed, namely, clicks/views, which is a percentage and reflects the attention degree of the certain content on the website page.

Conversion rate: the method is characterized in that in a statistical period, the ratio of the times of completing conversion behaviors to the total clicks of the promotion information is defined as orders/clicks.

The ES is an open source search engine developed based on Java, is designed for cloud computing, can achieve real-time search, and is stable, reliable and rapid.

Fig. 1 schematically shows an environment architecture diagram of an embodiment of the present application, as shown in the drawing:

the client 300 is connected to the server 100 through the network 200. The user inputs a search term through the client 300, so that the server 100 can obtain the corresponding search term. After obtaining the search terms, the server 100, in addition to obtaining normal search results according to the search terms, also rewrites the search terms according to a preset rewrite word table to obtain a plurality of rewritten search terms, wherein the rewrite word table is obtained according to historical search terms of a plurality of users in the same session, and includes a plurality of groups of rewrite word pairs for rewriting the search terms according to the rewrite word pairs therein to obtain rewritten search terms; after obtaining the rewritten search terms, the server 100 may determine the associated goods recommended to the user by using the rewritten search terms according to a preset search model; and returning the associated goods to the client 100 through the network 200 to be displayed in the same search page with the normal search result. The client 300 displays the associated goods, which may be displayed at the bottom of the search (feedblast).

In an exemplary embodiment, the server 100 may index a data center, such as a single house, or be distributed over different geographic locations (e.g., over several houses). The server 100 may provide services through one or more networks 200.

Network 200 includes various network devices such as routers, switches, multiplexers, hubs, modems, bridges, repeaters, firewalls, proxy devices, and/or the like. Network 200 may include physical links such as coaxial cable links, twisted pair cable links, fiber optic links, combinations thereof, and/or the like. The network 200 may include wireless links, such as cellular links, satellite links, wi-Fi links, and/or the like.

The client 300 may include a device such as a mobile device, a tablet device, a laptop computer, a smart device (e.g., smart apparel, smart watch, smart glasses), a virtual reality headset, a gaming device, a set-top box, a digital streaming device, a robot, a vehicle terminal, a smart television, a television box, or an e-book reader.

In the related technology, the commodities recommended at the bottom of the search are determined according to the factors of the user and the commodities and are irrelevant to the search terms of the user, so that the commodities recommended at the bottom of the search do not meet the requirements of the user, the interest of the user in clicking or purchasing cannot be aroused, and the conversion rate is low.

According to the searching scheme, the commodities recommended at the bottom of the search can meet the requirements of the user, the interest of the user in clicking or purchasing is improved, and the conversion rate is improved.

The searching scheme of the embodiment of the present application will be described in several embodiments, and for ease of understanding, the server 100 in fig. 1 will be exemplarily described as an executing body.

Example one

Fig. 2 schematically shows a flowchart of a search method according to an embodiment of the present application, which includes steps S410 to S430, and specifically describes the following:

in step S410, a first search term input by a target user is obtained.

The first search word refers to an original search word input in the client 300 by the target user. After the target user inputs the first search term at the client 300 and clicks on a search, the first search term is sent to the server 100 through the network 200, and the server 100 can obtain the first search term.

And step S420, rewriting the first search word according to a preset rewriting word table to obtain N second search words, wherein the rewriting word table is obtained according to historical search words of a plurality of users in the same session, and N is a positive integer.

The historical search terms in the same session refer to the historical search terms in a session (session). The one-time session may be defined as a session in a case where the one-time session is not left from the search page for more than a predetermined time or does not exit from searching the corresponding application, and the historical search term refers to a term that the user used to search once.

The server 100 may store the historical search terms of the users in the same session, so that mining may be performed according to the historical search terms of a plurality of (a large number of) users in the same session, thereby obtaining the rewritten vocabulary. The rewriting word list comprises a plurality of groups of rewriting word pairs and is used for rewriting the first search word according to the rewriting word pairs.

Since the historical search terms of the user in the same session generally have certain relevance, mining the historical search terms can obtain a corresponding rewritten word list. When mining the historical search words of the user in the same session, the server 100 may regard all the historical search words in the same session as having relevance, and respectively form a word pair in the rewrite word list by each historical search word and the remaining historical search words; or the historical search words in the same session may be evaluated, the word pairs meeting certain conditions are used as the word pairs in the rewritten word list, and the specific mining mode may be set according to actual needs, which is not specifically limited here.

After obtaining the first search word, the server 100 may match the first search word with a word in the rewritten word list, and rewrite the first search word according to a word corresponding to the word in the rewritten word list, which matches the first search word, so as to obtain a rewritten second search word. For example, if the first search term is a and the rewrite word list includes a rewrite word pair of a and B, then: and matching the search word A with the rewrite word pair in the rewrite word list, determining that the matched rewrite word pair is A and B, rewriting A into B, and thus obtaining B as a second search word. If the rewrite word table includes a plurality of rewrite word pairs that can be matched with the first search word, a plurality of second search words can be obtained.

And step S430, determining the associated commodities recommended to the target user according to the second search term, and returning the associated commodities.

When determining the associated goods recommended to the target user according to the second search term, the server 100 may perform operations such as recall, rough ranking, fine ranking and the like on the goods according to the second search term through a preset search model, so as to determine the associated goods recommended to the target user. The search model may include a recall sub-model, a rough-ranking sub-model, and a fine-ranking sub-model, so that the search model may complete corresponding operations of recall, rough-ranking, fine-ranking, and the like, and finally determine the associated goods returned to the target user.

It is understood that the server 100 may also perform operations such as recalling, coarse ranking, fine ranking and the like on the goods according to the first search term by using the search model, so as to determine the goods corresponding to the first search term directly input by the target user. Optionally, when determining the related goods recommended to the target user according to the second search term, the server 100 may perform independent sorting independent of the search corresponding to the first search term. And when the associated goods are returned, the search results corresponding to the first search terms can be returned together and displayed in different areas or modules in the same search page.

Referring to fig. 3, which is an application example diagram of the search method in the embodiment of the present application, as shown in the figure, the request may be split on line and divided into two groups, one is an experimental group and the search method in the embodiment of the present application is used, and the other is a control group and the current method recommended based on factors of the user and the commodity is used, so that the optimization effect of the search method in the embodiment of the present application may be determined by comparing the two groups. Specifically, if the traffic is the traffic of the control group, the original search bottom (feedblast) can be directly returned; if the flow rate is the flow rate of the experimental group, preprocessing such as rewriting and the search words input by the user can be carried out according to the mined rewriting word list, and finally the related commodities recommended to the user are determined to be displayed at the bottom of the search through recalling and sequencing. The rewritten word list can be mined in an off-line stage and synchronized to a DB (database) of the server side through hive; in the on-line stage, the rewritten word list can be synchronized to the memory, thereby facilitating the use of the line.

According to the searching method, a first search word input by a target user is obtained, the first search word is rewritten according to a preset rewriting word list, and N second search words are obtained, wherein the rewriting word list is obtained according to historical search words of a plurality of users in the same session; determining the associated commodities recommended to the target user according to the second search terms, and returning the associated commodities; because the rewriting vocabulary is obtained through the historical search words of a plurality of users in the same session, and the historical search words of the users in the same session have correlation, the original search words input by the users are rewritten through the rewriting vocabulary, and then the associated commodities are obtained according to the rewritten search words and recommended to the users, so that the commodities recommended at the bottom of the search (fed blast) and the search words of the users have certain correlation, the recommended associated commodities meet the requirements of the users, the interest of clicking or purchasing by the users is easily caused, the conversion rate is improved, and the purposes or effects of improving the number of payment users and the like are conveniently achieved.

In an exemplary embodiment, as shown in fig. 4, the searching method may further include steps S440 to S460, which are specifically as follows:

step S440, obtaining first historical search terms and second historical search terms of a plurality of users in the same session, where the first historical search terms are historical search terms whose corresponding click rates satisfy a first preset condition or whose corresponding conversion rates satisfy a second preset condition, and the second historical search terms are historical search terms whose corresponding click rates do not satisfy the first preset condition and whose conversion rates do not satisfy the second preset condition.

The first preset condition may include, but is not limited to, a click-through rate being greater than or equal to a click-through rate threshold. Likewise, the second preset condition may include, but is not limited to, the conversion being greater than or equal to the conversion threshold. Optionally, the first preset condition or the second preset condition may further define the historical search terms themselves, for example, the click rate of the historical search terms under sufficient exposure (i.e., the exposure is greater than a certain value) within a certain time (e.g., within 30 days) is greater than or equal to a preset click rate threshold. The more a search term is used by a user, the greater its exposure rate, and the less it is used, the smaller its exposure rate. Since search terms used by a small number of users have less representative meaning, selecting a history search term with sufficient exposure may make the determined first history search term and second search term more representative and more valuable to use.

The server 100 may obtain historical search terms of a plurality of users in the same session, compare the historical search terms in the same session with a first preset condition and a second preset condition, and determine that the historical search terms are the first historical search terms if a click rate corresponding to the historical search terms meets the first preset condition or a conversion rate corresponding to the historical search terms meets the second preset condition; and if the click rate corresponding to the historical search word does not meet the first preset condition and the corresponding conversion rate does not meet the second preset condition, determining that the historical search word is a second historical search word.

Step S450, the first historical search word and the second historical search word are respectively formed into word pairs to form a word pair set, and the word pair set comprises a plurality of groups of word pairs.

Specifically, each first historical search word and each second historical search word may respectively form a word pair, and then a word pair set composed of a plurality of sets of word pairs may be formed according to the formed word pairs. For example, if a first historical search term of a user in the same session is a, and second historical search terms are B and C, two word pairs of AB and AC may be formed. Alternatively, the first historical search word and the second historical search word may be randomly combined into a word pair, for example, in the foregoing example, an AB word pair or an AC word pair may be randomly generated.

It should be understood that there may be a plurality of history search terms in the same session, or there may be only one history search term, and when there is only one history search term, the history search term associated with the history search term cannot be determined, so in practical applications, if the first history search term and the second history search term of the same session are not obtained, the process of this mining may be ended.

Step S460, determining whether the word pairs in the word pair set satisfy a third preset condition, and storing the current word pair in the rewritten word list under the condition that the current word pair satisfies the third preset condition, where the third preset condition at least includes that the co-occurrence frequency of the word pair is greater than or equal to a frequency threshold.

When the third preset condition is that the co-occurrence frequency of the word pair is greater than or equal to the frequency threshold, the server 100 may calculate the co-occurrence frequency of each word pair in the word pair set, and compare the calculated result with the frequency threshold, thereby determining whether the co-occurrence frequency of each word pair is greater than or equal to the frequency threshold. For example, if the third preset condition is that the co-occurrence frequency of the word pair is greater than or equal to 20, the server 100 may compare the co-occurrence frequency of each word pair with 20, and if the co-occurrence frequency of each word pair is greater than or equal to 20, determine that the word pair satisfies the third preset condition, otherwise, the word pair does not satisfy the third preset condition. Of course, if the third preset condition further includes other conditions, the server 100 may determine whether the word pair satisfies the other conditions, and further determine whether the word pair satisfies the third preset condition. Under the condition that the current word pair meets a third preset condition, saving the current word pair into a rewriting word list to serve as one of the rewriting word pairs in the rewriting word list; and if the current word pair does not meet the third preset condition, abandoning the current word pair, and continuously determining whether other word pairs meet the third preset condition.

In the embodiment, first history search words and second history search words of a plurality of users in the same session are obtained, wherein the first history search words are history search words of which the corresponding click rates meet a first preset condition or of which the corresponding conversion rates meet a second preset condition, and the second history search words are history search words of which the corresponding click rates do not meet a first preset condition and the corresponding conversion rates do not meet a second preset condition; respectively forming word pairs by each first historical search word and each second historical search word to form a word pair set, determining whether the word pairs in the word pair set meet a third preset condition, and storing the current word pairs into a rewritten word list under the condition that the current word pairs meet the third preset condition, wherein the third preset condition at least comprises that the co-occurrence frequency of the word pairs is greater than or equal to a frequency threshold value; because the first historical search word is a historical search word with better click rate or conversion rate, and the second historical search word is a historical search word with poorer click rate or conversion rate, the first historical search word and the second historical search word form a word pair, and a rewriting word list is further determined, so that when the original search word (first search word) of the user is rewritten through the rewriting word list, a rewriting word (second search word) with better click rate or conversion rate can be obtained, and the click rate or conversion rate of the related goods is improved; in addition, the third preset condition at least comprises that the co-occurrence frequency of the word pairs is greater than or equal to the frequency threshold, and the co-occurrence frequency of the word pairs can effectively represent the relevance of the historical search words in the word pairs, so that the word pairs meeting the third preset condition are used as the rewrite words in the rewrite word list, and the relevant historical search words can be effectively used as the rewrite words.

In an exemplary embodiment, the third preset condition further includes that the similarity of the word pair is greater than or equal to a similarity threshold.

The similarity threshold may be set according to actual needs, for example, 0.6, and is not limited herein.

It is understood that the rewrite word list is determined by the co-occurrence frequency of word pairs, and although the history search word having a correlation can be determined as a rewrite word, the accuracy is low. The third preset condition also comprises that the similarity of the word pair is greater than or equal to a similarity threshold, the rewritten words can be determined through double standards, the rewriting precision of the rewritten word list is improved, and the determined associated commodities can better meet the requirements of target users.

In an exemplary embodiment, before step S460, that is, before determining whether the word pairs in the word pair set satisfy the third preset condition, as shown in fig. 5, steps S470 to S480 may be included, specifically as follows:

in step S470, semantic vectors of the first historical search term and the second historical search term in the word pair are obtained.

Optionally, when semantic vectors of a first historical search word and a second historical search word in the word pair are obtained, a word2vec model may be used for obtaining, where the word2vec model is a lightweight neural network structure and includes only an input layer, a hidden layer, and an output layer. The semantic vectors of the first historical search word and the second historical search word in the word pair are obtained by adopting a word2vec model, and the corresponding semantic vectors can be obtained by adopting a skip-gram algorithm in the word2vec model. Because the word2vec model is light, the similarity of the word pairs is determined by using the word2vec model, and the occupation of the computing resources of the server side can be reduced.

Step S480, determining the similarity of the first historical search word and the second historical search word in the word pair according to the semantic vector.

After obtaining the corresponding semantic vector, the server 100 may determine a similarity between the first historical search term and the second historical search term in the word pair according to a similarity algorithm. The similarity algorithm may be set according to actual needs, such as cosine similarity, and is not limited herein.

Fig. 6 is a diagram illustrating an application example of performing rewrite word mining by the search method according to the embodiment of the present application. As shown in the figure, the conversation may be divided into a conversation with a single search term (query) and a conversation with multiple search terms (query), and then the search terms are further divided into a first history search term (i.e., "good" query in the figure) and a second history search term (i.e., "bad" query in the figure) according to indexes such as click rate or conversion rate; and matching with the 'good' query and the 'bad' query based on the correlation relationship of the search words in the same session, and finally determining the rewritten word list according to the co-occurrence frequency and the similarity threshold of the matched word pairs. The similarity is obtained by obtaining a semantic vector through a word2vec model and then calculating the similarity of the cosine.

In this embodiment, by obtaining semantic vectors of the first historical search word and the second historical search word in the word pair, and determining the similarity between the first historical search word and the second historical search word in the word pair according to the semantic vectors, it may be convenient to determine whether the word pair satisfies a third preset condition.

In the related art, a synonym mining method is used for rewriting the search terms of the user, but this method requires two classifications, and at least ten thousand levels of labeling are manually performed to obtain training samples, so that a large amount of labor cost is required. In the embodiment, the word pairs consisting of the historical search words in the same session are determined through the co-occurrence frequency and the similarity of the word pairs, so that the automatic mining of the rewritten words can be realized, the rewritten and expanded recalling can be performed under the condition of no label, a large amount of labor cost is not required, and the implementation difficulty of the rewritten and expanded recalling is reduced. Meanwhile, the rewritten words can be subjected to quality labeling subsequently according to the actual on-line effect (such as conversion rate) and the like, so that the rewritten words in the rewritten word list can be further optimized.

In an exemplary embodiment, in step S430, the related product recommended to the target user is determined according to the second search term, and as shown in fig. 7, the method may include steps S431 to S433, which are as follows:

step S431, dividing the N second search terms into a first priority queue and a second priority queue, where the first priority queue and the second priority queue respectively include a plurality of second search terms.

When the N second search terms are divided into the first priority queue and the second priority queue, the word pairs may be sorted according to co-occurrence frequency and/or similarity, the first priority queue is placed before the sorting, the second priority queue is placed after the sorting, and the specific allocation rule may be set according to actual needs, which is not limited specifically here. The first priority queue and the second priority queue may include equal or unequal numbers of second search terms, for example, if there are 20 second search terms in total, the first 2 ordered second search terms may be placed in the first priority queue, and the remaining 18 second search terms may be placed in the second priority queue.

It will be appreciated that the number of second search terms is typically large in order to ensure that the search returns a large number of items. If all the second search terms are put into the search engine for searching, more computing resources of the search engine are occupied, and the performance of the search engine is influenced; and the second search terms are divided into two priority queues for searching, so that the influence on the performance of a search engine can be reduced while the quantity of commodities returned by searching is ensured. In practical applications, the N second search terms may be divided into a plurality of priority queues.

And step S432, recalling the first commodity according to the recall sub-model in the search model and the second search terms in the first priority queue.

Specifically, the first commodity is recalled according to a recall submodel in the search model and the second search term in the first priority queue, which may be that the first commodity is recalled according to the second search term in the first priority queue by using the recall submodel in the search model.

The search model may specifically include links such as recall, rough ranking, fine ranking and the like, so that recommendation of the commodity can be achieved, wherein the search model and a recall submodel, a rough ranking submodel and a fine ranking submodel included in the search model, and the rough ranking submodel may specifically adopt a manual rule strategy model, a linear model, a double-tower inner product depth model or a lightweight MLP (multi-layer perceptron) rough ranking model; the fine ranking sub-model can specifically adopt a linear model and can also adopt a deep learning model. The models and algorithms specifically adopted by the rough ranking sub model and the fine ranking sub model can be set according to actual needs, and are not specifically limited here.

The first commodity is a commodity which is recalled by the recall submodel according to the second search term in the first priority queue.

Step S433, determining the related commodities according to the search model and the first commodities under the condition that the quantity of the first commodities is larger than or equal to a quantity threshold value or the current request concurrency number is larger than or equal to a concurrency threshold value.

Wherein determining the associated item according to the search model and the first item may be: and determining the associated commodity according to the first commodity by adopting a search model. The number threshold and the concurrency threshold may be set according to actual needs, and are not particularly limited herein. When the number of the first commodities is larger than or equal to the number threshold, the number of the recalled commodities is enough, the return of a sufficient number of associated commodities can be guaranteed, and at the moment, the search is not required to be carried out according to the second search terms of the second priority queue, so that the occupation of search engine resources is further reduced. And when the current request concurrency number is greater than or equal to the concurrency threshold, the current request number is excessive (such as during activity), if the request is performed according to the second search word of the second priority queue, the request concurrency number is more, and occupied resources of the search engine are excessive, so that the associated commodities are determined directly according to the first commodities, and the occupation of the search engine resources can be reduced.

When the search model is used for determining the associated commodity according to the first commodity, the first commodity can be further screened by using a coarse ranking sub-model and a fine ranking sub-model in the search model, so that the final associated commodity is obtained.

In the embodiment, the N second search terms are divided into the first priority queue and the second priority queue, so that the influence on the performance of a search engine can be reduced while the quantity of returned commodities is ensured; and recalling the first commodities according to the recalling sub-model in the search model and the second search terms in the first priority queue, and determining associated commodities according to the search model and the first commodities under the condition that the quantity of the first commodities is greater than or equal to a quantity threshold value or the current request concurrency number is greater than or equal to a concurrency threshold value, so that the occupation of search engine resources can be further reduced, and the service performance of a search engine is improved.

In an exemplary embodiment, as shown in fig. 8, the searching method may further include steps S434 to S435, which are as follows:

in step S434, when the number of the first commodities is less than the number threshold and the number of requests is less than the preset number, a recall sub-model is used to recall the second commodities according to the second search term in the second priority queue.

The second commodity is a commodity which is recalled by the recall submodel according to the second search term in the second priority queue.

The preset times can be set according to actual needs, and are not limited here. Since the second search term is divided into two priority queues (the first priority queue and the second priority queue), and the search term in the same priority queue does not change the general search result when searching, the preset number of times can be consistent with the number of the priority queues. For example, if the number of the priority queues is two, the preset number of times may be 2. In some scenarios, the number of requests may be request batches, i.e., the requests belong to several request batches. For example, the request batch corresponding to the first priority queue is a first batch of requests, and the request batch corresponding to the second priority queue is a second batch of requests.

When the number of the first commodities is smaller than the number threshold, it indicates that the number of the recalled commodities is small and insufficient to return a sufficient number of associated commodities, and therefore, the server 100 may recall the commodities continuously by using the recall sub-model according to the second search word in the second priority queue to obtain a recalled second commodity when the number of requests is smaller than the preset number.

Step S435, determining a related item according to the search model, the first item, and the second item.

Specifically, step S435 may be: and determining the associated commodity according to the first commodity and the second commodity by adopting a search model. After recalling the first commodity and the second commodity, the server 100 may use the first commodity and the second commodity together as a recalled commodity according to the search model, and process the recalled commodities (the first commodity and the second commodity) through the rough-row sub model and the fine-row sub model, so as to obtain corresponding associated commodities.

In this embodiment, when the number of the first commodities is smaller than the number threshold and the number of requests is smaller than the preset number, the recall sub-model is used to recall the second commodities according to the second search term in the second priority queue, and then the associated commodities are determined according to the search model, the first commodities and the second commodities, so that the second search terms of the two priority queues can be used for searching, thereby ensuring that a large number of associated commodities are returned.

In an exemplary embodiment, the search method may further include: under the condition that the number of the second search words in the first priority queue is smaller than the first number, acquiring relevant words of the second search words in the first priority queue from the knowledge graph to perform candidate selection; or, under the condition that the number of the second search words in the second priority queue is smaller than the second number, acquiring relevant words of the second search words in the second priority queue from the knowledge graph to perform candidate selection.

The related words may include, but are not limited to, hypernyms, hyponyms, hypernyms, synonyms, or the like.

The first number and the second number may be set according to actual needs, for example, the first number may be 10, and the second number is 100, which is not limited herein. When the server 100 selects the relevant words from the knowledge graph for candidate, the number of the second search words in the first priority queue after candidate may be equal to the first number, and the number of the second search words in the second priority queue may be equal to the second number.

Because the second search terms in the first priority queue or the second search terms in the second priority queue may have a situation of insufficient quantity, the number of the returned associated commodities is small, and therefore, the second search terms can be guaranteed to have a certain number by acquiring the associated terms from the knowledge graph for candidate, so that the situation that more associated commodities are returned for the user to select is guaranteed, and the user experience is improved.

In an exemplary embodiment, as shown in fig. 9, in step S430, determining the associated product recommended to the target user according to the second search term may further include steps S510 to S550, which are specifically as follows:

and step S510, acquiring a third commodity output by the first sequencing submodel in the search model by adopting the second sequencing submodel in the search model, wherein the third commodity is a commodity determined by the search model according to the second search word.

The second ordering submodel may be the fine ordering submodel, and the first ordering submodel may be the coarse ordering submodel.

Specifically, the search model may be a product recalled according to the second search term through the recall sub-model, after passing through the rough ranking sub-model, a certain number of third products obtained after rough ranking are output, and the input of the fine ranking sub-model is the certain number of third products.

And step S520, scoring the third commodity by adopting the second sequencing submodel to obtain the sequencing score of the third commodity.

Step S530, respectively obtaining similarity between the first search term and the second search term corresponding to the third item, and determining a similarity score of the third item according to the similarity between the first search term and the second search term.

And step S540, determining the final score of the third commodity according to the sorting score and the similarity score.

And step S550, taking the third commodity with the final score larger than the score threshold value as a related commodity.

It can be understood that, if the associated product is determined directly according to the score of the fine ranking sub-model in the search model for the product, the associated product is actually determined by the second search term (i.e., the rewritten search term), and the used second search terms have different correlations with the first search term, so that while the product is scored by using the fine ranking sub-model, the score is comprehensively scored according to the similarity score obtained according to the similarity between the second search term corresponding to the product and the first search term, so that the correlation between the finally returned associated product and the original search term of the user is better, and the accuracy of recommending the associated product is further improved.

Please refer to fig. 10, which is a flowchart illustrating a searching method according to an embodiment of the present application. As shown in the figure, the second search term is divided into two priority queues, the first priority queue is used for requesting, if the result of the request is greater than or equal to 300 (namely the quantity threshold), or the number of the requests in concurrency during activity is too large (namely the number threshold is greater than the concurrency threshold), the second priority queue is not used for requesting, and the final associated commodity return is obtained directly according to the first priority queue through rough arrangement and fine arrangement; if the request result is less than 300 and the request frequency is less than 2 (corresponding to the preset frequency), continuing the request by adopting a second priority queue; after the second priority queue is adopted for requesting, because the request times at this time are equal to 2, the request circulation is not carried out any more, the coarse arrangement and the fine arrangement are carried out according to the current recalled commodities, and finally the associated commodities recommended to the user are determined to be returned. In the fine ranking stage, the final score of the commodity can be determined by combining the fine ranking score and the similarity, and then the related commodity is determined according to the final score.

Please refer to fig. 11, which is an exemplary diagram of a search result page obtained by the search method according to the embodiment of the present application. As shown in the figure, when the user searches for "dunhuang flying sky", the search result card display can also display other commodities related to the dunhuang series.

Example two

Fig. 12 schematically shows a block diagram of a search apparatus 600 according to the second embodiment of the present application, where the search apparatus 600 may be divided into one or more program modules, and the one or more program modules are stored in a storage medium and executed by one or more processors to complete the second embodiment of the present application. The program modules referred to in the embodiments of the present application refer to a series of computer program instruction segments that can perform specific functions, and the following description will specifically describe the functions of the program modules in the embodiments.

As shown in fig. 12, the search apparatus 600 may include an obtaining module 610, a rewriting module 620, and a determining module 630.

An obtaining module 610, configured to obtain a first search term input by a target user;

the rewriting module 620 is configured to rewrite the first search word according to a preset rewriting word list to obtain N second search words, where the rewriting word list is obtained according to historical search words of multiple users in the same session, and N is a positive integer;

and the determining module 630 is configured to determine the related goods recommended to the target user according to the second search term, and return the related goods.

In an exemplary embodiment, the search apparatus 600 further comprises a mining module (not shown in the figure), wherein the mining module is configured to: the method comprises the steps of obtaining first historical search words and second historical search words of a plurality of users in the same conversation, wherein the first historical search words are historical search words of which the corresponding click rates meet a first preset condition or the corresponding conversion rates meet a second preset condition, and the second historical search words are historical search words of which the corresponding click rates do not meet the first preset condition and the corresponding conversion rates do not meet the second preset condition; forming a word pair set by combining each first historical search word and each second historical search word respectively, wherein the word pair set comprises a plurality of groups of word pairs; and determining whether the word pairs in the word pair set meet a third preset condition, and storing the current word pairs into a rewritten word list under the condition that the current word pairs meet the third preset condition, wherein the third preset condition at least comprises that the co-occurrence frequency of the word pairs is greater than or equal to a frequency threshold value.

In an exemplary embodiment, the mining module is further configured to: obtaining semantic vectors of a first historical search word and a second historical search word in a word pair; and determining the similarity of the first historical search word and the second historical search word in the word pair according to the semantic vector.

In an exemplary embodiment, the determining module 630 is further configured to: dividing the N second search terms into a first priority queue and a second priority queue, wherein the first priority queue and the second priority queue respectively comprise a plurality of second search terms; recalling the first commodity according to the recall submodel in the search model and the second search term in the first priority queue; and determining the related commodities according to the search model and the first commodities under the condition that the quantity of the first commodities is greater than or equal to a quantity threshold value or the current request concurrency number is greater than or equal to a concurrency threshold value.

In an exemplary embodiment, the determining module 630 is further configured to: under the condition that the number of the first commodities is smaller than the number threshold value and the request times are smaller than the preset times, a recall sub-model is adopted to recall the second commodities according to the second search terms in the second priority queue; and determining the associated commodity according to the search model, the first commodity and the second commodity.

In an exemplary embodiment, the rewrite module 620 is further configured to: under the condition that the number of the second search words in the first priority queue is smaller than the first number, acquiring relevant words of the second search words in the first priority queue from the knowledge graph to perform candidate selection; or, under the condition that the number of the second search words in the second priority queue is smaller than the second number, acquiring relevant words of the second search words in the second priority queue from the knowledge graph to perform candidate selection.

In an exemplary embodiment, the determining module 630 is further configured to: acquiring a third commodity output by the first sequencing submodel in the search model by adopting a second sequencing submodel in the search model, wherein the third commodity is a commodity determined by the search model according to a second search word; grading the third commodity by adopting a second sequencing submodel to obtain a sequencing score of the third commodity; respectively obtaining the similarity between a second search word corresponding to a third commodity and the first search word, and determining the similarity score of the third commodity according to the similarity between the second search word and the first search word; determining a final score of the third commodity according to the ranking score and the similarity score; and taking the third commodity with the final score larger than the score threshold value as a related commodity.

EXAMPLE III

Fig. 13 is a hardware architecture diagram schematically illustrating a computer device 700 suitable for the search method according to the third embodiment of the present application. The computer device 700 may be a device capable of automatically performing numerical calculations and/or data processing according to instructions set or stored in advance. For example, the server may be a rack server, a blade server, a tower server or a rack server (including an independent server or a server cluster composed of a plurality of servers), a gateway, and the like. As shown in fig. 13, the computer device 700 includes at least, but is not limited to: memory 710, processor 720, and network interface 730 may be communicatively linked to each other by a system bus. Wherein:

the memory 710 includes at least one type of computer-readable storage medium including a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., SD or DX memory, etc.), a Random Access Memory (RAM), a Static Random Access Memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, an optical disk, etc. In some embodiments, the storage 710 may be an internal storage module of the computer device 700, such as a hard disk or a memory of the computer device 700. In other embodiments, the memory 710 may also be an external storage device of the computer device 700, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), or the like, provided on the computer device 700. Of course, the memory 710 may also include both internal and external memory modules of the computer device 700. In this embodiment, the memory 710 is generally used for storing an operating system and various application software installed in the computer device 700, such as program codes of a search method. In addition, the memory 710 may also be used to temporarily store various types of data that have been output or are to be output.

Processor 720 may be a Central Processing Unit (CPU), controller, microcontroller, microprocessor, or other data Processing chip in some embodiments. The processor 720 generally serves to control overall operation of the computer device 700, such as to perform control and processing related to data interaction or communication with the computer device 700. In this embodiment, processor 720 is configured to execute program codes stored in memory 710 or to process data.

The network interface 730 may include a wireless network interface or a wired network interface, and the network interface 730 is typically used to establish communication links between the computer device 700 and other computer devices. For example, the network interface 730 is used to connect the computer device 700 to an external terminal via a network, establish a data transmission channel and a communication link between the computer device 700 and the external terminal, and the like. The network may be a wireless or wired network such as an Intranet (Intranet), the Internet (Internet), a Global System of Mobile communication (GSM), wideband Code Division Multiple Access (WCDMA), a 4G network, a 5G network, bluetooth (Bluetooth), or Wi-Fi.

It is noted that fig. 13 only shows a computer device having components 710-730, but it is understood that not all of the shown components are required to be implemented, and that more or fewer components may be implemented instead.

In this embodiment, the search method stored in the memory 710 may be further divided into one or more program modules and executed by one or more processors (in this embodiment, the processor 720) to implement the embodiments of the present application.

Example four

Embodiments of the present application also provide a computer-readable storage medium, on which a computer program is stored, and the computer program, when executed by a processor, implements the steps of the search method in the embodiments.

In this embodiment, the computer-readable storage medium includes a flash memory, a hard disk, a multimedia card, a card type memory (e.g., SD or DX memory, etc.), a Random Access Memory (RAM), a Static Random Access Memory (SRAM), a Read Only Memory (ROM), an Electrically Erasable Programmable Read Only Memory (EEPROM), a Programmable Read Only Memory (PROM), a magnetic memory, a magnetic disk, an optical disk, and the like. In some embodiments, the computer readable storage medium may be an internal storage unit of the computer device, such as a hard disk or a memory of the computer device. In other embodiments, the computer readable storage medium may be an external storage device of the computer device, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), and the like provided on the computer device. Of course, the computer-readable storage medium may also include both internal and external storage devices of the computer device. In this embodiment, the computer-readable storage medium is generally used for storing an operating system and various types of application software installed in the computer device, for example, the program codes of the search method in the embodiment, and the like. Further, the computer-readable storage medium may also be used to temporarily store various types of data that have been output or are to be output.

It will be apparent to those skilled in the art that the modules or steps of the embodiments of the present application described above may be implemented by a general purpose computing device, they may be centralized on a single computing device or distributed across a network of multiple computing devices, and alternatively, they may be implemented by program code executable by a computing device, such that they may be stored in a storage device and executed by a computing device, and in some cases, the steps shown or described may be performed in an order different from that described herein, or they may be separately fabricated into individual integrated circuit modules, or multiple ones of them may be fabricated into a single integrated circuit module. Thus, embodiments of the present application are not limited to any specific combination of hardware and software.

The above description is only a preferred embodiment of the present application, and not intended to limit the scope of the present application, and all the equivalent structures or equivalent processes that can be directly or indirectly applied to other related technical fields by using the contents of the specification and the drawings of the present application are also included in the scope of the present application.

Claims

1. A method of searching, comprising:

acquiring a first search term input by a target user;

rewriting the first search word according to a preset rewriting word table to obtain N second search words, wherein the rewriting word table is obtained according to historical search words of a plurality of users in the same session, and N is a positive integer;

and determining the associated commodity recommended to the target user according to the second search word, and returning the associated commodity.

2. The search method of claim 1, further comprising:

acquiring first historical search terms and second historical search terms of a plurality of users in the same session, wherein the first historical search terms are historical search terms of which the corresponding click rates meet a first preset condition or the corresponding conversion rates meet a second preset condition, and the second historical search terms are historical search terms of which the corresponding click rates do not meet the first preset condition and the corresponding conversion rates do not meet the second preset condition;

respectively forming word pairs by the first historical search word and the second historical search word to form a word pair set, wherein the word pair set comprises a plurality of groups of word pairs;

and determining whether the word pairs in the word pair set meet a third preset condition, and saving the current word pairs to the rewritten word list under the condition that the current word pairs meet the third preset condition, wherein the third preset condition at least comprises that the co-occurrence frequency of the word pairs is greater than or equal to a frequency threshold value.

3. The method according to claim 2, wherein the third preset condition further includes that a similarity of the word pair is greater than or equal to a similarity threshold.

4. The searching method according to claim 3, prior to said determining whether the word pairs in the set of word pairs satisfy a third preset condition, further comprising:

obtaining semantic vectors of the first historical search word and the second historical search word in the word pair;

determining similarity of the first historical search word and the second historical search word in the word pair according to the semantic vector.

5. The searching method according to any one of claims 1 to 4, wherein the determining of the associated commodity recommended to the target user according to the second search term comprises:

dividing the N second search terms into a first priority queue and a second priority queue, wherein the first priority queue and the second priority queue respectively comprise a plurality of second search terms;

recalling the first commodity according to a recall submodel in the search model and the second search terms in the first priority queue;

and determining the associated commodity according to the search model and the first commodity when the quantity of the first commodity is greater than or equal to a quantity threshold value or the current request concurrency number is greater than or equal to a concurrency threshold value.

6. The search method of claim 5, further comprising:

under the condition that the number of the first commodities is smaller than the number threshold value and the request times are smaller than preset times, recalling second commodities by adopting the recall sub-model according to the second search terms in the second priority queue;

determining the associated commodity according to the search model, the first commodity and the second commodity.

7. The search method of claim 5, further comprising:

under the condition that the number of the second search words in the first priority queue is smaller than a first number, acquiring relevant words of the second search words in the first priority queue from a knowledge graph to perform candidate selection; or,

and under the condition that the number of the second search words in the second priority queue is smaller than a second number, acquiring relevant words of the second search words in the second priority queue from the knowledge graph to perform candidate selection.

8. The searching method according to any one of claims 1 to 4, wherein the determining of the associated commodity recommended to the target user according to the second search term comprises:

acquiring a third commodity output by the first sequencing submodel in the search model by adopting a second sequencing submodel in the search model, wherein the third commodity is a commodity determined by the search model according to the second search word;

grading the third commodity by adopting the second sequencing submodel to obtain a sequencing score of the third commodity;

respectively obtaining the similarity between a second search word corresponding to the third commodity and the first search word, and determining the similarity score of the third commodity according to the similarity between the second search word and the first search word;

determining a final score of the third commodity according to the ranking score and the similarity score;

and taking the third commodity with the final score larger than a score threshold value as the associated commodity.

9. A search apparatus, comprising:

the acquisition module is used for acquiring a first search word input by a target user;

the rewriting module is used for rewriting the first search word according to a preset rewriting word list to obtain N second search words, the rewriting word list is obtained according to historical search words of a plurality of users in the same session, and N is a positive integer;

and the determining module is used for determining the associated commodities recommended to the target user according to the second search terms and returning the associated commodities.

10. A computer arrangement comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor, when executing the computer program, is adapted to carry out the steps of the search method according to any of claims 1 to 8.

11. A computer-readable storage medium, in which a computer program is stored which is executable by at least one processor to cause the at least one processor to perform the steps of the search method of any one of claims 1 to 8.