A method, device, computer equipment and storage medium for determining recommended words

Abstract

The present disclosure provides a method, apparatus, computer device and storage medium for determining a recommendation word, wherein the method comprises: in response to receiving search information, selecting a first search result that meets a first interaction requirement from the search results corresponding to the search information; inputting the search information and the first search result into a pre-trained first neural network to obtain a first feature vector corresponding to the search information; determining a target recommendation word from each candidate recommendation word based on the first feature vector and a second feature vector corresponding to each candidate recommendation word; wherein the second feature vector is obtained after inputting the candidate recommendation word and a second search result that meets the second interaction requirement and corresponds to the candidate recommendation word into a pre-trained second neural network; the target recommendation word is used for display on a search results page.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a method, device, computer equipment and storage medium for determining a recommendation word. Background Art

[0002] Search is an important way for users to obtain information within an application. Users can enter search terms in the application's search bar, and the application's search engine can return document content related to the user's search terms. In order to improve the user's search efficiency, in addition to displaying search results that match the search information on the search results page, recommended terms related to the search information can also be displayed.

[0003] In the related art, when determining recommended words related to search information, text relevance algorithms such as the Best Matching 25 (BM25) algorithm and the Term Frequency-Inverse Document Frequency (TF-IDF) algorithm are often used to determine recommended words that contain the same content as the search information input by the user. The recommended words obtained by this method can only be recommended from the text level of the search information, so that the recommended words recommended to the user may not meet the user's needs for searching again. Summary of the invention

[0004] The embodiments of the present disclosure at least provide a method, apparatus, computer device, and storage medium for determining a recommendation word.

[0005] In a first aspect, an embodiment of the present disclosure provides a method for determining a recommendation word, comprising:

[0006] In response to receiving the search information, selecting a first search result that meets the first interaction requirement from the search results corresponding to the search information;

[0007] Inputting the search information and the first search result into a pre-trained first neural network to obtain a first feature vector corresponding to the search information;

[0008] Based on the first feature vector and the second feature vectors corresponding to each candidate recommendation word, a target recommendation word is determined from the candidate recommendation words; wherein the second feature vector is obtained after inputting the candidate recommendation word and the second search result corresponding to the candidate recommendation word and meeting the second interaction requirement into a pre-trained second neural network; the target recommendation word is used for display on the search results page.

[0009] In a possible implementation manner, the first neural network and the second neural network are trained according to the following steps:

[0010] Obtaining sample data and a sample label corresponding to the sample data; wherein the sample label corresponding to the sample data is used to indicate whether the sample data is positive sample data;

[0011] Inputting the first sample search information in the sample data and the first sample search result that satisfies the third interaction requirement corresponding to the first sample search information into the first neural network to be trained, so as to obtain a first sample vector corresponding to the first sample search information; and inputting the second sample search information in the sample data and the second sample search result that satisfies the fourth interaction requirement corresponding to the second sample search information into the second neural network to be trained, so as to obtain a second sample vector corresponding to the second sample search information;

[0012] Based on the first sample vector, the second sample vector and the sample label, a loss value of this training is determined, and network parameters of the first neural network and the second neural network are adjusted based on the loss value.

[0013] In a possible implementation manner, the sample data includes positive sample data and negative sample data, wherein:

[0014] The positive sample data includes at least one first search information group that meets a preset condition in the same search sequence; wherein the search sequence includes search information corresponding to at least two search requests initiated by the same user terminal, and the time interval between the search requests meets the preset time interval requirement;

[0015] The negative sample data includes at least one second search information group that does not meet the preset condition in the same search sequence; and / or at least one third search information group in a different search sequence;

[0016] Among them, the first search information group and the second search information group include first sample search information input by a user in the same search sequence; the first search information group also includes second sample search information corresponding to a third search result that meets the preset interaction requirements; the second search information group also includes second sample search information corresponding to a fourth search result that does not meet the preset interaction requirements; and the third search information group contains multiple search information located in different search sequences.

[0017] In a possible implementation manner, inputting the search information and the first search result into a pre-trained first neural network to obtain a first feature vector corresponding to the search information includes:

[0018] Concatenate the search information and key information of the first search result to obtain concatenated first input information;

[0019] Performing feature encoding processing on the first input information to obtain target input information corresponding to the first input information;

[0020] The first input information is input into a pre-trained first neural network to obtain a first feature vector output by the first neural network and corresponding to the search information.

[0021] In a possible implementation manner, the step of concatenating the search information and the key information of the first search result to obtain the concatenated first input information includes:

[0022] Performing word segmentation processing on the search information according to a plurality of preset word segmentation granularities to obtain word segmentation results corresponding to the search information at different word segmentation granularities;

[0023] The word segmentation results corresponding to the search information at different word segmentation granularities and the key information of the first search result are spliced ​​in a preset splicing order to obtain the spliced ​​first input information.

[0024] In a possible implementation manner, the second feature vector is stored in an approximate nearest neighbor index library;

[0025] The determining a target recommendation word from each candidate recommendation word based on the first feature vector and the second feature vectors respectively corresponding to each candidate recommendation word includes:

[0026] Determine, from the second feature vectors corresponding to the candidate recommendation words stored in the approximate nearest neighbor, a target feature vector whose similarity with the first feature vector meets a preset similarity requirement;

[0027] The candidate recommendation word corresponding to the target feature vector is the target recommendation word.

[0028] In a possible implementation manner, the first interaction requirement and the second interaction requirement include feature thresholds and / or quantity ranking requirements under multiple interaction attribute features;

[0029] The multiple interactive attribute features include the following:

[0030] The popularity information of search results, the number of times search results are displayed, and the number of times search results are triggered.

[0031] In a second aspect, the present disclosure also provides a device for determining a recommendation word, including:

[0032] A selection module, configured to select, in response to receiving the search information, a first search result that meets the first interaction requirement from the search results corresponding to the search information;

[0033] An input module, configured to input the search information and the first search result into a pre-trained first neural network to obtain a first feature vector corresponding to the search information;

[0034] A determination module is used to determine a target recommendation word from each candidate recommendation word based on the first feature vector and the second feature vector corresponding to each candidate recommendation word; wherein the second feature vector is obtained after inputting the candidate recommendation word and the second search result corresponding to the candidate recommendation word and meeting the second interaction requirement into a pre-trained second neural network; the target recommendation word is used to be displayed on the search results page.

[0035] In a possible implementation manner, the determination module is further configured to train the first neural network and the second neural network according to the following steps:

[0036] Obtaining sample data and a sample label corresponding to the sample data; wherein the sample label corresponding to the sample data is used to indicate whether the sample data is positive sample data;

[0037] Inputting the first sample search information in the sample data and the first sample search result that satisfies the third interaction requirement corresponding to the first sample search information into the first neural network to be trained, so as to obtain a first sample vector corresponding to the first sample search information; and inputting the second sample search information in the sample data and the second sample search result that satisfies the fourth interaction requirement corresponding to the second sample search information into the second neural network to be trained, so as to obtain a second sample vector corresponding to the second sample search information;

[0038] Based on the first sample vector, the second sample vector and the sample label, a loss value of this training is determined, and network parameters of the first neural network and the second neural network are adjusted based on the loss value.

[0039] In a possible implementation manner, the sample data includes positive sample data and negative sample data, wherein:

[0040] The positive sample data includes at least one first search information group that meets a preset condition in the same search sequence; wherein the search sequence includes search information corresponding to at least two search requests initiated by the same user terminal, and the time interval between the search requests meets the preset time interval requirement;

[0041] The negative sample data includes at least one second search information group that does not meet the preset condition in the same search sequence; and / or at least one third search information group in a different search sequence;

[0042] Among them, the first search information group and the second search information group include first sample search information input by a user in the same search sequence; the first search information group also includes second sample search information corresponding to a third search result that meets the preset interaction requirements; the second search information group also includes second sample search information corresponding to a fourth search result that does not meet the preset interaction requirements; and the third search information group contains multiple search information located in different search sequences.

[0043] In a possible implementation manner, when the input module inputs the search information and the first search result into a pre-trained first neural network to obtain a first feature vector corresponding to the search information, it is used to:

[0044] Concatenate the search information and key information of the first search result to obtain concatenated first input information;

[0045] Performing feature encoding processing on the first input information to obtain target input information corresponding to the first input information;

[0046] The first input information is input into a pre-trained first neural network to obtain a first feature vector output by the first neural network and corresponding to the search information.

[0047] In a possible implementation manner, when the input module performs splicing processing on the search information and the key information of the first search result to obtain the spliced ​​first input information, it is used to:

[0048] Performing word segmentation processing on the search information according to a plurality of preset word segmentation granularities to obtain word segmentation results corresponding to the search information at different word segmentation granularities;

[0049] The word segmentation results corresponding to the search information at different word segmentation granularities and the key information of the first search result are spliced ​​in a preset splicing order to obtain the spliced ​​first input information.

[0050] In a possible implementation manner, the second feature vector is stored in an approximate nearest neighbor index library;

[0051] The determination module, when determining a target recommendation word from each candidate recommendation word based on the first feature vector and the second feature vectors respectively corresponding to each candidate recommendation word, is used to:

[0052] Determine, from the second feature vectors corresponding to the candidate recommendation words stored in the approximate nearest neighbor, a target feature vector whose similarity with the first feature vector meets a preset similarity requirement;

[0053] The candidate recommendation word corresponding to the target feature vector is the target recommendation word.

[0054] In a possible implementation manner, the first interaction requirement and the second interaction requirement include feature thresholds and / or quantity ranking requirements under multiple interaction attribute features;

[0055] The multiple interactive attribute features include the following:

[0056] The popularity information of search results, the number of times search results are displayed, and the number of times search results are triggered.

[0057] In a third aspect, an embodiment of the present disclosure further provides a computer device, comprising: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processor and the memory communicate via the bus, and when the machine-readable instructions are executed by the processor, the steps of the above-mentioned first aspect, or any possible implementation of the first aspect are performed.

[0058] In a fourth aspect, an embodiment of the present disclosure further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned first aspect or any possible implementation of the first aspect are executed.

[0059] The method, device, computer equipment and storage medium for determining recommended words provided by the embodiments of the present disclosure can select a first search result that meets the first interaction requirement from the search results corresponding to the search information, and input the search information and the first search result into a pre-trained first neural network to obtain a first feature vector corresponding to the search information; then, based on the first feature vector and the second feature vectors corresponding to the candidate recommended words, determine the target recommended word from the candidate recommended words. In this way, since the features of the search results are integrated when determining the first feature vector and the second feature vector, the semantic expression of the first feature vector and the second feature vector is richer, and the target recommended words determined based on the first feature vector and the second feature vector have a better semantic generalization effect; further, since the first search result and the second search result are search results that meet the first interaction requirement or the second interaction requirement, the search results of the target recommended words determined based on the first feature vector and the second feature vector can better meet the user's search needs, thereby improving the user's search efficiency.

[0060] In order to make the above-mentioned objectives, features and advantages of the present disclosure more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following briefly introduces the drawings required for use in the embodiments. The drawings herein are incorporated into the specification and constitute a part of the specification. These drawings illustrate embodiments that meet the present disclosure and are used together with the specification to illustrate the technical solutions of the present disclosure. It should be understood that the following drawings only illustrate certain embodiments of the present disclosure and should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can also be obtained based on these drawings without creative work.

[0062] Figure 1 A flow chart showing a method for determining a recommendation word provided by an embodiment of the present disclosure is shown;

[0063] Figure 2 A schematic diagram of a search result page in the method for determining a recommendation word provided in an embodiment of the present disclosure is shown;

[0064] Figure 3 A schematic diagram showing the architecture of a device for determining a recommendation word provided by an embodiment of the present disclosure;

[0065] Figure 4 A schematic diagram of the structure of a computer device provided by an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical scheme and advantages of the embodiments of the present disclosure clearer, the technical scheme in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, rather than all of the embodiments. The components of the embodiments of the present disclosure generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present disclosure provided in the drawings is not intended to limit the scope of the present disclosure for protection, but merely represents the selected embodiments of the present disclosure. Based on the embodiments of the present disclosure, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present disclosure.

[0067] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0068] The term "and / or" herein only describes an association relationship, indicating that three relationships may exist. For example, A and / or B may represent the following three situations: A exists alone, A and B exist at the same time, and B exists alone. In addition, the term "at least one" herein represents any combination of at least two of any one or more of a plurality of. For example, including at least one of A, B, and C may represent including any one or more elements selected from the set consisting of A, B, and C.

[0069] It is understandable that before using the technical solutions disclosed in the embodiments of the present disclosure, the types, scope of use, usage scenarios, etc. of the personal information involved in the present disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0070] For example, in response to receiving an active request from a user, a prompt message is sent to the user to clearly prompt the user that the operation requested to be performed will require obtaining and using the user's personal information. Thus, the user can autonomously choose whether to provide personal information to software or hardware such as an electronic device, application, server, or storage medium that performs the operation of the technical solution of the present disclosure according to the prompt message.

[0071] As an optional but non-limiting implementation, in response to receiving an active request from the user, the prompt information may be sent to the user in the form of a pop-up window, in which the prompt information may be presented in text form. In addition, the pop-up window may also carry a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0072] It is understandable that the above notification and the process of obtaining user authorization are merely illustrative and do not constitute a limitation on the implementation of the present disclosure. Other methods that meet the relevant laws and regulations may also be applied to the implementation of the present disclosure.

[0073] Through research, it is found that when determining the recommended words related to the search information, taking the search information "I am reborn" as an example, the recommended words corresponding to the search information may be "The night of rebirth XXXX", "After rebirth XXXX", "I am XXXX", etc., which are recommended words containing part of the search content in the search information. Since the recommended words are similar to the search results of the search information, if the user cannot find the search results they want to browse in the search results corresponding to the current search information, when the user initiates a search again based on the recommended words, they still may not find the search results they want to browse. Therefore, the effect of guiding the user's search is limited, and the search results displayed after the search is re-initiated may not be the content that the user is interested in.

[0074] Based on the above research, the present disclosure provides a method, device, computer equipment and storage medium for determining recommendation words, which can select a first search result that meets the first interaction requirement from the search results corresponding to the search information, and input the search information and the first search result into a pre-trained first neural network to obtain a first feature vector corresponding to the search information; then, based on the first feature vector and the second feature vectors corresponding to each candidate recommendation word, determine the target recommendation word from each candidate recommendation word. In this way, since the features of the search results are integrated when determining the first feature vector and the second feature vector, the semantic expression of the first feature vector and the second feature vector is richer, and the target recommendation word determined based on the first feature vector and the second feature vector has a better semantic generalization effect; further, since the first search result and the second search result are search results that meet the first interaction requirement or the second interaction requirement, the search results of the target recommendation word determined based on the first feature vector and the second feature vector can better meet the user's search needs, thereby improving the user's search efficiency.

[0075] To facilitate understanding of this embodiment, a method for determining a recommendation word disclosed in an embodiment of the present disclosure is first introduced in detail. The execution subject of the method for determining a recommendation word provided in an embodiment of the present disclosure is generally a computer device with certain computing capabilities, and the computer device includes, for example: a terminal device or a server or other processing device, and the terminal device can be a user equipment (User Equipment, UE), a mobile device, a user terminal, a terminal, a personal digital assistant (Personal Digital Assistant, PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. In some possible implementations, the method for determining a recommendation word can be implemented by a processor calling a computer-readable instruction stored in a memory.

[0076] See also Figure 1 FIG. 1 is a flowchart of a method for determining a recommendation word provided by an embodiment of the present disclosure, wherein the method includes S101 to S103, wherein:

[0077] S101: In response to receiving search information, selecting a first search result that meets a first interaction requirement from search results corresponding to the search information.

[0078] S102: Input the search information and the first search result into a pre-trained first neural network to obtain a first feature vector corresponding to the search information.

[0079] S103: Determine a target recommendation word from the candidate recommendation words based on the first feature vector and the second feature vectors corresponding to the candidate recommendation words respectively; wherein the second feature vector is obtained by inputting the candidate recommendation word and the second search result corresponding to the candidate recommendation word and meeting the second interaction requirement into a pre-trained second neural network; the target recommendation word is used for display on the search results page.

[0080] The following is a detailed description of the above steps.

[0081] For S101, the first interaction requirement may include feature thresholds and / or quantity sorting requirements under multiple interaction attribute features, and the multiple interaction attribute features may include popularity information of search results, number of times search results are displayed, number of times search results are triggered, etc.; wherein, the quantity sorting requirement is used to characterize the ranking requirement in the sorting result corresponding to the interaction attribute feature after sorting according to any of the interaction attribute features. The quantity sorting requirement may, for example, be the top 5 in the sorting result after sorting from high to low according to the number of times the search results are displayed.

[0082] Specifically, when selecting the first search result that meets the first interaction requirement from the search results corresponding to the search information, the first search result that meets the filtering conditions corresponding to each feature threshold can be selected from the search results corresponding to the search information according to the feature thresholds under each interaction attribute feature; and / or, the first search result that meets each quantity sorting requirement can be selected from the search results corresponding to the search information according to the quantity sorting requirement under each interaction attribute feature.

[0083] Exemplarily, taking the example that the interactive attribute characteristics include the number of times the search results are displayed and the number of times they are triggered, when selecting the first search result, the search results can be sorted from high to low according to the number of times the search results are displayed and the number of times they are triggered, respectively, to generate a first queue corresponding to the number of times the display is performed and a second queue corresponding to the number of times the trigger is performed, and the top five search results are selected from the first queue and the second queue as the first search result, respectively.

[0084] S102: Input the search information and the first search result into a pre-trained first neural network to obtain a first feature vector corresponding to the search information.

[0085] Here, the first neural network may be a network capable of deep learning, and the network type of the first neural network may be, for example, a fully connected neural network.

[0086] In a possible implementation manner, when obtaining the first feature vector corresponding to the search information, the following steps A1 to A3 may be performed:

[0087] A1: Concatenate the search information and key information of the first search result to obtain concatenated first input information.

[0088] Here, the key information of the first search result may include title information, summary information, etc. of the first search result.

[0089] In a possible implementation manner, when the search information and the key information of the first search result are concatenated to obtain the concatenated first input information, the following steps A11 to A12 may be performed:

[0090] A11: Perform word segmentation processing on the search information according to a plurality of preset word segmentation granularities to obtain word segmentation results corresponding to the search information at different word segmentation granularities.

[0091] Here, the word segmentation granularity may include 1 word, 2 words, etc., and the word segmentation granularity specifically included in the multiple word segmentation granularities may be set according to actual needs, for example, may include 1 word and 2 words.

[0092] Exemplarily, taking the search information "super power city" as an example, according to the unigram model, "super power city" is segmented with a granularity of 1 word, and the segmentation results "super", "power", "city", and "city" can be obtained; according to the bigram model, "super power city" is segmented with a granularity of 2 words, and the segmentation results "super power", "power city", and "city" can be obtained; according to the trigram model, "super power city" is segmented with a granularity of 3 words, and the segmentation results "super power city" and "power city" can be obtained.

[0093] In this way, by performing word segmentation processing on the search information according to a plurality of preset word segmentation granularities, a word segmentation result for representing the semantics of the search information in different dimensions can be obtained, which is convenient for the subsequent first neural network to perform semantic feature extraction.

[0094] A12: The word segmentation results corresponding to the search information at different word segmentation granularities and the key information of the first search result are spliced ​​in a preset splicing order to obtain the spliced ​​first input information.

[0095] In this way, by splicing the word segmentation results corresponding to the search information at different word segmentation granularities and the key information of the first search result according to a preset splicing order, the first input information containing features representing different semantic dimensions of the search information and features under the search result dimension can be obtained.

[0096] A2: Perform feature encoding processing on the first input information to obtain target input information corresponding to the first input information.

[0097] Here, when performing feature encoding processing on the first input information, an embedding layer embedding can be used to perform feature encoding processing on the first input information to convert the first input information into target input information that meets the input requirements of the first neural network.

[0098] A3: Input the first input information into a pre-trained first neural network to obtain a first feature vector output by the first neural network that corresponds to the search information.

[0099] S103: Determine a target recommendation word from the candidate recommendation words based on the first feature vector and the second feature vectors corresponding to the candidate recommendation words respectively; wherein the second feature vector is obtained by inputting the candidate recommendation word and the second search result corresponding to the candidate recommendation word and meeting the second interaction requirement into a pre-trained second neural network; the target recommendation word is used for display on the search results page.

[0100] Here, the candidate recommendation words may be recommendation words set in advance according to a specific application scenario. Taking the application scenario of novel search as an example, the candidate recommendation words may be composed of the names of all books in the online library. That is, assuming that there are 100,000 books in the online library, the names corresponding to the 100,000 books respectively constitute the candidate recommendation words; the second feature vector may be stored in an approximate nearest neighbor index library (ANN), so as to achieve fast search of the second feature vector; the second interaction requirement may include feature thresholds and / or quantity sorting requirements under a variety of interaction attribute features, and the various interaction attribute features include the popularity information of the search results, the number of times the search results are displayed, the number of times the search results are triggered, etc.

[0101] Exemplarily, the schematic diagram of the search results page may be as follows: Figure 2 As shown, Figure 2 The search information "Rebirth of I am", the search result "Rebirth of I am X", the search result "Rebirth of I am XX" and the target recommendation words "Rebirth in another world XXXX, XXX after rebirth, universal rebirth XXXX, road to rebirth XXX" are displayed.

[0102] In one possible implementation, when determining a target recommendation word from the candidate recommendation words, a target feature vector whose similarity with the first feature vector meets a preset similarity requirement can be determined from the second feature vectors corresponding to the candidate recommendation words respectively stored in the approximate nearest neighbor; the candidate recommendation word corresponding to the target feature vector is the target recommendation word.

[0103] Here, the preset similarity requirement may, for example, be that a similarity score used to characterize the degree of similarity with the first feature vector is greater than a preset similarity threshold; and / or, a score sorting requirement corresponding to the similarity score satisfies a preset score sorting requirement, and the score sorting requirement may, for example, be that after sorting the similarity scores of each second feature vector with the first feature vector from high to low, the second feature vectors are located in the top 8 positions in the sorting result.

[0104] Specifically, the similarity score can be obtained by, for example, multiplying the first feature vector and the second feature vector; or, the familiarity score can also be obtained by calculating the cosine similarity between the first feature vector and the second feature vector.

[0105] It should be noted that all parameters that can characterize the similarity between the first eigenvector and the second eigenvector can be used as a basis for calculating the similarity score, and the embodiments of the present disclosure do not limit which parameters and methods are used to calculate the similarity score.

[0106] Furthermore, in order to avoid the target recommendation words displayed in the search results page being the same as the search results corresponding to the search information, after the target recommendation words are determined, the target recommendation words can also be filtered according to the search results corresponding to the search information, and target recommendation words different from the search results corresponding to the search information can be displayed in the search results page; in order to ensure that the number of target recommendation words displayed in the search results page is fixed, when it is detected that the target recommendation words are the same as the search results corresponding to the search information, the target number of target recommendation words can be re-determined based on the target number of target recommendation words detected to be the same as the search results corresponding to the search information.

[0107] In a possible implementation manner, when training the first neural network and the second neural network, the following steps B1 to B3 may be performed:

[0108] B1: Obtain sample data and a sample label corresponding to the sample data; wherein the sample label corresponding to the sample data is used to indicate whether the sample data is positive sample data.

[0109] Here, the sample data may include positive sample data and negative sample data, wherein:

[0110] The positive sample data may include at least one first search information group that meets a preset condition in the same search sequence; wherein the search sequence includes search information corresponding to at least two search requests initiated by the same user terminal, and the time interval between the search requests meets the preset time interval requirement;

[0111] The negative sample data may include at least one second search information group that does not meet the preset condition in the same search sequence; and / or at least one third search information group in different search sequences.

[0112] Here, the search sequence may refer to a sequence of search requests whose time intervals between adjacent search requests meet the preset time interval requirement. The search sequence may include search information of at least one search request, and the above-mentioned positive sample data and negative sample data may be a search sequence from search information containing at least two search requests.

[0113] Specifically, the first search information group and the second search information group include first sample search information input by a user in the same search sequence; the first search information group also includes second sample search information corresponding to a third search result that meets preset interaction requirements; the second search information group also includes second sample search information corresponding to a fourth search result that does not meet the preset interaction requirements; and the third search information group includes multiple search information located in different search sequences.

[0114] Among them, the time interval between each search request in the search sequence meets the preset time interval requirement, for example, the time interval between each search request can be less than 15 minutes; the preset interaction requirement may include that the display time of the search result details page corresponding to the search result in the user terminal exceeds the preset time, and / or, the target button corresponding to the search result in the user terminal is triggered, and the target button includes a favorite button, a share button, etc.

[0115] For example, user A enters the search information "Reborn as I am" in the search box of application 1, and triggers the search result "Reborn as I am XXX" after initiating the search request, and browses the search result details page displayed after the trigger; user A initiates a search for "Reborn in Another World XX" 10 minutes after searching for "Reborn as I am", and triggers the search result "Reborn in Another World XXXXXX" after initiating the search request, and browses the search result details page displayed after the trigger; user A initiates a search for "Super Power City XXX" 5 minutes after searching for "Reborn in Another World XX", and then exits application 1. The search sequence corresponding to user A's search this time can be search information "Reborn as I am"-search result "Reborn as I am XXX"-search information "Reborn in Another World XX"-search result "Reborn in Another World XXXXXX"-search information "Super Power City XXX", so that the user search behavior recorded after the user authorization can be recorded.

[0116] Among them, the first sample search information that can be used as the positive sample data is the search information "Reborn as XXX" input by the user, and the second sample search information that can be used as the positive sample data is the search information "Reborn in Another World XX" whose corresponding search results meet the preset interaction requirements. The positive sample data can be "Reborn as XXX"-"Reborn in Another World XX", and the sample label corresponding to the positive sample data can be 1, which is used to characterize the type of sample data as positive sample data; the first sample search information that can be used as the negative sample data is the search information "Reborn as XXX" input by the user, and the second sample search information that can be used as the negative sample data is the search information "Super Power City XXX" whose corresponding search results do not meet the preset interaction requirements. The negative sample data can be "Reborn as XXX"-"Super Power City XXX", and the sample label corresponding to the positive sample data can be 0, which is used to characterize the type of sample data as negative sample data.

[0117] B2: Input the first sample search information in the sample data and the first sample search result that meets the third interaction requirement corresponding to the first sample search information into the first neural network to be trained to obtain the first sample vector corresponding to the first sample search information; and input the second sample search information in the sample data and the second sample search result that meets the fourth interaction requirement corresponding to the second sample search information into the second neural network to be trained to obtain the second sample vector corresponding to the second sample search information.

[0118] Here, the third interaction requirement may include feature thresholds and / or quantity ranking requirements under multiple interaction attribute features, and the multiple interaction attribute features include popularity information of search results, number of times search results are displayed, number of times search results are triggered, etc.; the fourth interaction requirement may include feature thresholds under multiple interaction attribute features, and the multiple interaction attribute features include popularity information of search results, number of times search results are displayed, number of times search results are triggered, etc.

[0119] B3: Based on the first sample vector, the second sample vector and the sample label, determine the loss value of this training, and adjust the network parameters of the first neural network and the second neural network based on the loss value.

[0120] Here, when determining the loss value of this training, the similarity score corresponding to the sample data can be determined based on the first sample vector and the second sample vector, and the prediction result corresponding to the sample data can be determined based on the similarity score. The preset result is used to characterize whether the first sample search information and the second sample search information in the sample data are similar (that is, whether the second sample search information is the recommended word corresponding to the first sample search information). The loss value of this training is determined based on the preset target loss function, the prediction result corresponding to the sample data, and the sample label corresponding to the sample data.

[0121] The target loss function may be, for example, a cross entropy loss function. When adjusting the network parameters of the first neural network and the second neural network, the network parameters may be updated by using a gradient descent method.

[0122] The method for determining recommended words provided by the embodiment of the present disclosure can select a first search result that meets the first interaction requirement from the search results corresponding to the search information, and input the search information and the first search result into a pre-trained first neural network to obtain a first feature vector corresponding to the search information; then, based on the first feature vector and the second feature vectors corresponding to the candidate recommended words, determine the target recommended word from the candidate recommended words. In this way, since the features of the search results are integrated when determining the first feature vector and the second feature vector, the semantic expression of the first feature vector and the second feature vector is richer, and the target recommended words determined based on the first feature vector and the second feature vector have a better semantic generalization effect; further, since the first search result and the second search result are search results that meet the first interaction requirement or the second interaction requirement, the search results of the target recommended words determined based on the first feature vector and the second feature vector can better meet the user's search needs, thereby improving the user's search efficiency.

[0123] Those skilled in the art will appreciate that, in the above method of specific implementation, the order in which the steps are written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of the steps should be determined by their functions and possible internal logic.

[0124] Based on the same inventive concept, the embodiment of the present disclosure also provides a device for determining recommendation words corresponding to the method for determining recommendation words. Since the principle of solving the problem by the device in the embodiment of the present disclosure is similar to the method for determining recommendation words in the embodiment of the present disclosure, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0125] Reference Figure 3 , which is a schematic diagram of the architecture of a device for determining a recommendation word provided by an embodiment of the present disclosure, the device comprises: a selection module 301, an input module 302, and a determination module 303; wherein,

[0126] A selection module 301, configured to select, in response to receiving search information, a first search result that meets a first interaction requirement from search results corresponding to the search information;

[0127] An input module 302, configured to input the search information and the first search result into a pre-trained first neural network to obtain a first feature vector corresponding to the search information;

[0128] The determination module 303 is used to determine a target recommendation word from each candidate recommendation word based on the first feature vector and the second feature vector corresponding to each candidate recommendation word; wherein the second feature vector is obtained after inputting the candidate recommendation word and the second search result corresponding to the candidate recommendation word and meeting the second interaction requirement into a pre-trained second neural network; the target recommendation word is used to be displayed on the search results page.

[0129] In a possible implementation manner, the determining module 303 is further configured to train the first neural network and the second neural network according to the following steps:

[0130] Obtaining sample data and a sample label corresponding to the sample data; wherein the sample label corresponding to the sample data is used to indicate whether the sample data is positive sample data;

[0131] Inputting the first sample search information in the sample data and the first sample search result that satisfies the third interaction requirement corresponding to the first sample search information into the first neural network to be trained, so as to obtain a first sample vector corresponding to the first sample search information; and inputting the second sample search information in the sample data and the second sample search result that satisfies the fourth interaction requirement corresponding to the second sample search information into the second neural network to be trained, so as to obtain a second sample vector corresponding to the second sample search information;

[0132] Based on the first sample vector, the second sample vector and the sample label, a loss value of this training is determined, and network parameters of the first neural network and the second neural network are adjusted based on the loss value.

[0133] In a possible implementation manner, the sample data includes positive sample data and negative sample data, wherein:

[0134] The positive sample data includes at least one first search information group that meets a preset condition in the same search sequence; wherein the search sequence includes search information corresponding to at least two search requests initiated by the same user terminal, and the time interval between the search requests meets the preset time interval requirement;

[0135] The negative sample data includes at least one second search information group that does not meet the preset condition in the same search sequence; and / or at least one third search information group in a different search sequence;

[0136] Among them, the first search information group and the second search information group include first sample search information input by a user in the same search sequence; the first search information group also includes second sample search information corresponding to a third search result that meets the preset interaction requirements; the second search information group also includes second sample search information corresponding to a fourth search result that does not meet the preset interaction requirements; and the third search information group contains multiple search information located in different search sequences.

[0137] In a possible implementation manner, when the input module 302 inputs the search information and the first search result into a pre-trained first neural network to obtain a first feature vector corresponding to the search information, it is configured to:

[0138] Concatenate the search information and key information of the first search result to obtain concatenated first input information;

[0139] Performing feature encoding processing on the first input information to obtain target input information corresponding to the first input information;

[0140] The first input information is input into a pre-trained first neural network to obtain a first feature vector output by the first neural network and corresponding to the search information.

[0141] In a possible implementation manner, the input module 302, when concatenating the search information and the key information of the first search result to obtain the concatenated first input information, is used to:

[0142] Performing word segmentation processing on the search information according to a plurality of preset word segmentation granularities to obtain word segmentation results corresponding to the search information at different word segmentation granularities;

[0143] The word segmentation results corresponding to the search information at different word segmentation granularities and the key information of the first search result are spliced ​​in a preset splicing order to obtain the spliced ​​first input information.

[0144] In a possible implementation manner, the second feature vector is stored in an approximate nearest neighbor index library;

[0145] The determination module 303, when determining a target recommendation word from each candidate recommendation word based on the first feature vector and the second feature vectors respectively corresponding to each candidate recommendation word, is used to:

[0146] Determine, from the second feature vectors corresponding to the candidate recommendation words stored in the approximate nearest neighbor, a target feature vector whose similarity with the first feature vector meets a preset similarity requirement;

[0147] The candidate recommendation word corresponding to the target feature vector is the target recommendation word.

[0148] In a possible implementation manner, the first interaction requirement and the second interaction requirement include feature thresholds and / or quantity ranking requirements under multiple interaction attribute features;

[0149] The multiple interactive attribute features include the following:

[0150] The popularity information of search results, the number of times search results are displayed, and the number of times search results are triggered.

[0151] The device for determining recommended words provided by the embodiment of the present disclosure can select a first search result that meets the first interaction requirement from the search results corresponding to the search information, and input the search information and the first search result into a pre-trained first neural network to obtain a first feature vector corresponding to the search information; then, based on the first feature vector and the second feature vectors corresponding to the candidate recommended words, determine the target recommended word from the candidate recommended words. In this way, since the features of the search results are integrated when determining the first feature vector and the second feature vector, the semantic expressions of the first feature vector and the second feature vector are richer, and the target recommended words determined based on the first feature vector and the second feature vector have better semantic generalization effects; further, since the first search result and the second search result are search results that meet the first interaction requirement or the second interaction requirement, the search results of the target recommended words determined based on the first feature vector and the second feature vector can better meet the user's search needs, thereby improving the user's search efficiency.

[0152] For descriptions of the processing flow of each module in the device and the interaction flow between each module, reference may be made to the relevant descriptions in the above method embodiment, which will not be described in detail here.

[0153] Based on the same technical concept, the embodiment of the present disclosure also provides a computer device. Figure 4 As shown, it is a schematic diagram of the structure of a computer device 400 provided in an embodiment of the present disclosure, including a processor 401, a memory 402, and a bus 403. Among them, the memory 402 is used to store execution instructions, including a memory 4021 and an external memory 4022; the memory 4021 here is also called an internal memory, which is used to temporarily store the operation data in the processor 401, and the data exchanged with the external memory 4022 such as a hard disk. The processor 401 exchanges data with the external memory 4022 through the memory 4021. When the computer device 400 is running, the processor 401 communicates with the memory 402 through the bus 403, so that the processor 401 executes the following instructions:

[0154] In response to receiving the search information, selecting a first search result that meets the first interaction requirement from the search results corresponding to the search information;

[0155] Inputting the search information and the first search result into a pre-trained first neural network to obtain a first feature vector corresponding to the search information;

[0156] Based on the first feature vector and the second feature vectors corresponding to each candidate recommendation word, a target recommendation word is determined from the candidate recommendation words; wherein the second feature vector is obtained after inputting the candidate recommendation word and the second search result corresponding to the candidate recommendation word and meeting the second interaction requirement into a pre-trained second neural network; the target recommendation word is used for display on the search results page.

[0157] In a possible implementation manner, in the instructions of the processor 401, the first neural network and the second neural network are trained according to the following steps:

[0158] Obtaining sample data and a sample label corresponding to the sample data; wherein the sample label corresponding to the sample data is used to indicate whether the sample data is positive sample data;

[0159] Inputting the first sample search information in the sample data and the first sample search result that satisfies the third interaction requirement corresponding to the first sample search information into the first neural network to be trained, so as to obtain a first sample vector corresponding to the first sample search information; and inputting the second sample search information in the sample data and the second sample search result that satisfies the fourth interaction requirement corresponding to the second sample search information into the second neural network to be trained, so as to obtain a second sample vector corresponding to the second sample search information;

[0160] Based on the first sample vector, the second sample vector and the sample label, a loss value of this training is determined, and network parameters of the first neural network and the second neural network are adjusted based on the loss value.

[0161] In a possible implementation manner, in the instruction of the processor 401, the sample data includes positive sample data and negative sample data, wherein:

[0162] The positive sample data includes at least one first search information group that meets a preset condition in the same search sequence; wherein the search sequence includes search information corresponding to at least two search requests initiated by the same user terminal, and the time interval between the search requests meets the preset time interval requirement;

[0163] The negative sample data includes at least one second search information group that does not meet the preset condition in the same search sequence; and / or at least one third search information group in a different search sequence;

[0164] Among them, the first search information group and the second search information group include first sample search information input by a user in the same search sequence; the first search information group also includes second sample search information corresponding to a third search result that meets the preset interaction requirements; the second search information group also includes second sample search information corresponding to a fourth search result that does not meet the preset interaction requirements; and the third search information group contains multiple search information located in different search sequences.

[0165] In a possible implementation manner, in the instruction of the processor 401, inputting the search information and the first search result into a pre-trained first neural network to obtain a first feature vector corresponding to the search information includes:

[0166] Concatenate the search information and key information of the first search result to obtain concatenated first input information;

[0167] Performing feature encoding processing on the first input information to obtain target input information corresponding to the first input information;

[0168] The first input information is input into a pre-trained first neural network to obtain a first feature vector output by the first neural network and corresponding to the search information.

[0169] In a possible implementation manner, in the instruction of the processor 401, the step of concatenating the search information and key information of the first search result to obtain the concatenated first input information includes:

[0170] Performing word segmentation processing on the search information according to a plurality of preset word segmentation granularities to obtain word segmentation results corresponding to the search information at different word segmentation granularities;

[0171] The word segmentation results corresponding to the search information at different word segmentation granularities and the key information of the first search result are spliced ​​in a preset splicing order to obtain the spliced ​​first input information.

[0172] In a possible implementation manner, in the instruction of the processor 401, the second feature vector is stored in an approximate nearest neighbor index library;

[0173] The determining a target recommendation word from each candidate recommendation word based on the first feature vector and the second feature vectors respectively corresponding to each candidate recommendation word includes:

[0174] Determine, from the second feature vectors corresponding to the candidate recommendation words stored in the approximate nearest neighbor, a target feature vector whose similarity with the first feature vector meets a preset similarity requirement;

[0175] The candidate recommendation word corresponding to the target feature vector is the target recommendation word.

[0176] In a possible implementation manner, in the instruction of the processor 401, the first interaction requirement and the second interaction requirement include feature thresholds and / or quantity ranking requirements under multiple interaction attribute features;

[0177] The multiple interactive attribute features include the following:

[0178] The popularity information of search results, the number of times search results are displayed, and the number of times search results are triggered.

[0179] The present disclosure also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method for determining a recommendation word described in the above method embodiment are executed. The storage medium may be a volatile or non-volatile computer-readable storage medium.

[0180] The present disclosure also provides a computer program product that carries a program code. The program code includes instructions that can be used to execute the steps of the method for determining recommendation words described in the above method embodiment. For details, please refer to the above method embodiment, which will not be repeated here.

[0181] The computer program product may be implemented in hardware, software or a combination thereof. In one optional embodiment, the computer program product is implemented as a computer storage medium. In another optional embodiment, the computer program product is implemented as a software product, such as a software development kit (SDK).

[0182] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, the specific working process of the system and device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. In the several embodiments provided in the present disclosure, it should be understood that the disclosed system, device and method can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of the device or unit can be electrical, mechanical or other forms.

[0183] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0184] In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0185] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present disclosure, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present disclosure. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0186] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present disclosure, which are used to illustrate the technical solutions of the present disclosure, rather than to limit them. The protection scope of the present disclosure is not limited thereto. Although the present disclosure is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed in the present disclosure, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure shall be subject to the protection scope of the claims.

Claims

1. A method for determining a recommendation word, characterized in that: include: In response to receiving the search information, selecting a first search result that meets the first interaction requirement from the search results corresponding to the search information; Inputting the search information and the first search result into a pre-trained first neural network to obtain a first feature vector corresponding to the search information; Based on the first feature vector and the second feature vectors corresponding to the candidate recommendation words, a target recommendation word is determined from the candidate recommendation words; wherein the second feature vector is obtained by inputting the candidate recommendation word and the second search result corresponding to the candidate recommendation word and meeting the second interaction requirement into a pre-trained second neural network; the target recommendation word is used to be displayed on the search result page, The steps of training the first neural network and the second neural network include: Acquire sample data; wherein the sample data includes positive sample data and negative sample data, the positive sample data includes at least one first search information group that meets a preset condition in the same search sequence; the negative sample data includes at least one second search information group that does not meet the preset condition in the same search sequence, and / or at least one third search information group in a different search sequence; the search sequence is a sequence of search requests whose time intervals between adjacent search requests meet a preset time interval requirement; The search sequence includes search information corresponding to at least two search requests initiated by the same user terminal; Among them, the first search information group and the second search information group include first sample search information input by a user in the same search sequence; the first search information group also includes second sample search information corresponding to a third search result that meets the preset interaction requirements; the second search information group also includes second sample search information corresponding to a fourth search result that does not meet the preset interaction requirements; and the third search information group contains multiple search information located in different search sequences.

2. The method according to claim 1, characterized in that The step of training the first neural network and the second neural network also includes: Obtaining a sample label corresponding to the sample data; wherein the sample label corresponding to the sample data is used to indicate whether the sample data is positive sample data; Inputting the first sample search information in the sample data and the first sample search result that satisfies the third interaction requirement corresponding to the first sample search information into the first neural network to be trained, so as to obtain a first sample vector corresponding to the first sample search information; and inputting the second sample search information in the sample data and the second sample search result that satisfies the fourth interaction requirement corresponding to the second sample search information into the second neural network to be trained, so as to obtain a second sample vector corresponding to the second sample search information; Based on the first sample vector, the second sample vector and the sample label, a loss value of this training is determined, and network parameters of the first neural network and the second neural network are adjusted based on the loss value.

3. The method according to claim 1, characterized in that The step of inputting the search information and the first search result into a pre-trained first neural network to obtain a first feature vector corresponding to the search information includes: Concatenate the search information and key information of the first search result to obtain concatenated first input information; Performing feature encoding processing on the first input information to obtain target input information corresponding to the first input information; The target input information is input into a pre-trained first neural network to obtain a first feature vector output by the first neural network corresponding to the search information.

4. The method according to claim 3, characterized in that The step of concatenating the search information and the key information of the first search result to obtain the concatenated first input information includes: Performing word segmentation processing on the search information according to a plurality of preset word segmentation granularities to obtain word segmentation results corresponding to the search information at different word segmentation granularities; The word segmentation results corresponding to the search information at different word segmentation granularities and the key information of the first search result are spliced ​​in a preset splicing order to obtain the spliced ​​first input information.

5. The method according to claim 1, characterized in that The second feature vector is stored in an approximate nearest neighbor index library; The determining a target recommendation word from each candidate recommendation word based on the first feature vector and the second feature vectors respectively corresponding to each candidate recommendation word includes: Determine, from the second feature vectors corresponding to the candidate recommendation words stored in the approximate nearest neighbor index library, a target feature vector whose similarity with the first feature vector meets a preset similarity requirement; The candidate recommendation word corresponding to the target feature vector is the target recommendation word.

6. The method according to claim 1, characterized in that The first interaction requirement and the second interaction requirement include feature thresholds and / or quantity ranking requirements under multiple interaction attribute features; The multiple interactive attribute features include the following: The popularity information of search results, the number of times search results are displayed, and the number of times search results are triggered.

7. A device for determining a recommendation word, characterized in that: include: A selection module, configured to select, in response to receiving the search information, a first search result that meets the first interaction requirement from the search results corresponding to the search information; An input module, configured to input the search information and the first search result into a pre-trained first neural network to obtain a first feature vector corresponding to the search information; A determination module is used to determine a target recommendation word from each candidate recommendation word based on the first feature vector and a second feature vector corresponding to each candidate recommendation word; wherein the second feature vector is obtained by inputting the candidate recommendation word and a second search result corresponding to the candidate recommendation word and meeting a second interaction requirement into a pre-trained second neural network; the target recommendation word is used to be displayed on the search result page, The determining module is further used to train the first neural network and the second neural network according to the following steps: Acquire sample data; wherein the sample data includes positive sample data and negative sample data, the positive sample data includes at least one first search information group that meets a preset condition in the same search sequence; the negative sample data includes at least one second search information group that does not meet the preset condition in the same search sequence; and / or, at least one third search information group in a different search sequence; the search sequence is a sequence of search requests whose time intervals between adjacent search requests meet a preset time interval requirement; wherein the search sequence includes search information corresponding to at least two search requests initiated by the same user terminal; Among them, the first search information group and the second search information group include first sample search information input by a user in the same search sequence; the first search information group also includes second sample search information corresponding to a third search result that meets the preset interaction requirements; the second search information group also includes second sample search information corresponding to a fourth search result that does not meet the preset interaction requirements; and the third search information group contains multiple search information located in different search sequences.

8. A computer device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processor and the memory communicate via the bus, and when the machine-readable instructions are executed by the processor, the steps of the method for determining a recommendation word as described in any one of claims 1 to 6 are performed.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for determining a recommendation word according to any one of claims 1 to 6 are executed.

Images