Search intention recognition method and device, computer device and storage medium
By constructing a user behavior data framework and utilizing a deep neural network with neural networks and gated recurrent units, the problem of low accuracy in future intent recognition in traditional search intent recognition methods is solved, and high-accuracy prediction of users' future search intent is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- INDUSTRIAL AND COMMERCIAL BANK OF CHINA
- Filing Date
- 2022-08-30
- Publication Date
- 2026-05-22
AI Technical Summary
Traditional search intent recognition methods only identify the user's current search intent, resulting in low accuracy in recognizing the user's future search intent.
By acquiring multiple behavioral data points from users, performing feature adjustment processing, constructing a user behavior data framework, and utilizing an intent recognition model, including neural networks, update gates, and reset gates, to predict users' search intent, similar search information is merged to improve recognition accuracy.
It improves the accuracy of identifying users' future search intentions by constructing a user behavior data framework and using a deep neural network with gated recurrent units, thereby enabling the correlation processing and prediction of user behavior data.
Smart Images
Figure CN115408583B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of artificial intelligence technology, and in particular to a search intent recognition method, apparatus, computer device, and storage medium. Background Technology
[0002] With the rapid development of mobile internet technology and the market, various mobile applications (APPs) with large user bases and integrating multiple business modules have become important platforms for user traffic distribution. The global search mechanism commonly found in mobile APPs is a crucial means of user traffic distribution. Simultaneously, users generate a large amount of behavioral data on mobile devices, and user preferences change rapidly. Therefore, the ability to quickly identify users' short-term interests or intentions based on their recent behavior is vital for search engine search performance.
[0003] Traditional search intent recognition methods employ Neuro-Linguistic Programming (NLP) techniques combined with knowledge graph technology to deeply process users' search terms. Then, they utilize text classification and prediction models based on deep learning or machine learning to analyze users' search intent, aiming to provide users with more accurate and comprehensive search information. However, this method only identifies the user's current search intent, resulting in low accuracy in recognizing the user's future search intent. Summary of the Invention
[0004] Therefore, it is necessary to provide a search intent recognition method, apparatus, computer device, computer-readable storage medium, and computer program product to address the aforementioned technical problems.
[0005] Firstly, this application provides a method for identifying search intent. The method includes:
[0006] Obtain multiple behavioral data points from users;
[0007] Each behavioral data point is subjected to feature adjustment processing to obtain multiple behavioral feature data; the behavioral feature data is the characteristic data information of the behavioral data.
[0008] Based on the aforementioned behavioral feature data, a user behavior data framework is constructed; the user behavior data framework is a data framework that includes all behavioral feature data of the user, as well as the correlation relationships between the aforementioned behavioral feature data.
[0009] By using the intent recognition model and based on the user behavior data framework, the user's search intent is predicted to obtain the user's predicted search information.
[0010] Optionally, the step of constructing a user behavior data framework based on each of the behavioral feature data includes:
[0011] Input the behavioral feature data into the data framework model to obtain the initial user behavior data framework;
[0012] In the initial user behavior data framework, data association processing is performed on each of the behavioral feature data to obtain the user behavior data framework.
[0013] Optionally, the intent recognition model includes a neural network, an update gate, and a reset gate; the step of predicting and recognizing the user's search intent based on the user behavior data framework using the intent recognition model to obtain the user's predicted search information includes:
[0014] Obtain the user's current search information;
[0015] The neural network identifies the search intent of the user behavior data framework, and selects the current search information that is similar to the search intent from among the current search information as the first predicted search information.
[0016] Through the update gate, among the current search information other than each of the first predicted search information, the current search information that is similar to the search intent is selected as the second predicted search information;
[0017] The first predicted search information and the second predicted search information are calculated through the reset gate. If the correlation value is greater than a preset correlation threshold, the first predicted search information and the second predicted search information are merged to obtain the initial predicted search information.
[0018] If the preset iteration termination condition is not met, the process returns to identifying the search intent of the user behavior data framework through the neural network, and selecting current search information similar to the search intent from each current search information as the first predicted search information step, until the preset iteration termination condition is met, and then all the initial predicted search information is used as the user's predicted search information.
[0019] Optionally, the method further includes:
[0020] Obtain the behavioral feature data of each sample user, as well as the sample search information of the sample users;
[0021] The sample user's behavioral feature data and the sample user's search information are input into the initial intent recognition model, and the initial intent recognition model is trained to obtain the intent recognition model.
[0022] Optionally, after obtaining the intent recognition model, the method further includes:
[0023] Acquire the behavioral characteristic data of offline users, as well as the search information of the offline users;
[0024] Based on the behavioral feature data, the predicted search information of the offline user is determined using an intent recognition model.
[0025] If the similarity between the predicted search information of the offline user and the search information of the offline user is less than a preset similarity value, return to the step of obtaining the sample behavioral feature data of each sample user and the sample search information of the sample user, until the similarity between the predicted search information of the offline user and the search information of the offline user is not less than the preset similarity value, and output the intent recognition model obtained in the last iteration.
[0026] Optionally, after obtaining the user's predicted search information, the method further includes:
[0027] Obtain the user's current behavioral characteristics data and the user's search information;
[0028] The current behavioral feature data and the user's search information are input into the intent recognition model, and the intent recognition model is adjusted to obtain the adjusted intent recognition model.
[0029] Secondly, this application also provides a search intent recognition device. The device includes:
[0030] The acquisition module is used to acquire multiple behavioral data of the user;
[0031] The processing module is used to perform feature adjustment processing on each of the behavioral data to obtain multiple behavioral feature data; the behavioral feature data is the characteristic data information of the behavioral data;
[0032] A construction module is used to construct a user behavior data framework based on the behavioral feature data; the user behavior data framework is a data framework that includes all behavioral feature data of the user and the correlation between the behavioral feature data.
[0033] The prediction module is used to predict the user's search intent based on the user behavior data framework using an intent recognition model, thereby obtaining the user's predicted search information.
[0034] Optionally, the building module is specifically used for:
[0035] Input the behavioral feature data into the data framework model to obtain the initial user behavior data framework;
[0036] In the initial user behavior data framework, data association processing is performed on each of the behavioral feature data to obtain the user behavior data framework.
[0037] Optionally, the intent recognition model includes a neural network, an update gate, and a reset gate; the prediction module is specifically used for:
[0038] Obtain the user's current search information;
[0039] The neural network identifies the search intent of the user behavior data framework, and selects the current search information that is similar to the search intent from among the current search information as the first predicted search information.
[0040] Through the update gate, among the current search information other than each of the first predicted search information, the current search information that is similar to the search intent is selected as the second predicted search information;
[0041] The first predicted search information and the second predicted search information are calculated through the reset gate. If the correlation value is greater than a preset correlation threshold, the first predicted search information and the second predicted search information are merged to obtain the initial predicted search information.
[0042] If the preset iteration termination condition is not met, the process returns to identifying the search intent of the user behavior data framework through the neural network, and selecting current search information similar to the search intent from each current search information as the first predicted search information step, until the preset iteration termination condition is met, and then all the initial predicted search information is used as the user's predicted search information.
[0043] Optionally, the device further includes:
[0044] The sample acquisition module is used to acquire the sample behavioral feature data of each sample user, as well as the sample search information of the sample user;
[0045] The training module is used to input the sample behavioral feature data of each sample user and the sample search information of the sample user into the initial intent recognition model, and train the initial intent recognition model to obtain the intent recognition model.
[0046] Optionally, the device further includes:
[0047] The feature acquisition module is used to acquire various behavioral feature data of offline users, as well as the search information of the offline users;
[0048] The offline prediction module is used to determine the predicted search information of the offline user based on the behavioral feature data through an intent recognition model.
[0049] The optimization module is used to return to the step of obtaining the sample behavioral feature data of each sample user and the sample search information of the sample user when the similarity between the predicted search information of the offline user and the search information of the offline user is less than a preset similarity value, until the similarity between the predicted search information of the offline user and the search information of the offline user is not less than the preset similarity value, and then output the intent recognition model obtained in the last iteration.
[0050] Optionally, the device further includes:
[0051] The reacquisition module is used to acquire the user's current behavioral characteristic data and the user's search information;
[0052] The adjustment module is used to input the current behavior feature data and the user's search information into the intent recognition model, adjust the intent recognition model, and obtain the adjusted intent recognition model.
[0053] Thirdly, this application provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method described in any one of the first aspects.
[0054] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of the method described in any one of the first aspects.
[0055] Fifthly, this application provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps of the method described in any one of the first aspects.
[0056] The aforementioned search intent recognition method, apparatus, computer device, and storage medium acquire multiple user behavioral data; perform feature adjustment processing on each behavioral data to obtain multiple behavioral feature data; construct a user behavior data framework based on each behavioral feature data; the user behavior data framework is a data framework that includes all the user's behavioral feature data and the correlation relationships between each behavioral feature data; and predict the user's search intent using an intent recognition model based on the user behavior data framework to obtain the user's predicted search information. By performing feature processing on the user's behavioral data to construct a user behavior data framework and predicting the user's search information based on the user behavior data framework, the accuracy of recognizing the user's future search intent is improved. Attached Figure Description
[0057] Figure 1 This is a flowchart illustrating a search intent recognition method in one embodiment;
[0058] Figure 2 This is a flowchart illustrating the steps for predicting search information in one embodiment;
[0059] Figure 3 This is a schematic diagram of a deep neural network based on a controlled recurrent unit in one embodiment;
[0060] Figure 4 This is a flowchart illustrating a method for identifying the search intent of different users in one embodiment;
[0061] Figure 5 This is a schematic diagram of the process for obtaining training sample data in one embodiment;
[0062] Figure 6 This is a flowchart illustrating an example of search intent recognition in one embodiment;
[0063] Figure 7 This is a structural block diagram of a search intent recognition device in one embodiment;
[0064] Figure 8 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation
[0065] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0066] The search intent recognition method provided in this application can be applied to a terminal, a server, or a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. The terminal can include, but is not limited to, various personal computers, laptops, tablets, etc. This terminal performs feature processing on user behavior data to construct a user behavior data framework, and predicts user search information based on this framework, thereby improving the accuracy of recognizing future user search intent.
[0067] In one embodiment, such as Figure 1 As shown, a method for recognizing search intent is provided. Taking an application terminal of this method as an example, the method includes the following steps:
[0068] Step S101: Obtain multiple behavioral data of the user.
[0069] In this embodiment, with the user's authorization, the terminal obtains information such as the user's platform search history, browsed pages, browsing duration, and search methods, and uses the above information as the user's behavioral data.
[0070] Step S102: Perform feature adjustment processing on each behavioral data to obtain multiple behavioral feature data.
[0071] In this embodiment, the terminal uses behavioral feature adjustment technology to adjust the data features of each behavioral data point, obtaining multiple behavioral feature data points for subsequent intent recognition processing. The behavioral feature adjustment network can be, but is not limited to, a deep neural network based on gated recurrent units. For example, if a user's behavioral feature is searching for information via voice input, then the user behavioral feature data extracted by the behavioral feature adjustment network is the voice search data.
[0072] Step S103: Construct a user behavior data framework based on the behavioral feature data.
[0073] The user behavior data framework is a data framework that includes all user behavior feature data and the relationships between these behavioral feature data.
[0074] In this embodiment, the terminal establishes associations between various behavioral feature data and constructs a behavioral feature data framework based on all the established associations. The specific construction process will be explained in detail later.
[0075] Step S104: Using the intent recognition model and based on the user behavior data framework, predict the user's search intent to obtain the user's predicted search information.
[0076] In this embodiment, the terminal uses an intent recognition model to predict the user's search intent based on the user behavior data framework, and obtains the user's predicted search information.
[0077] Based on the above scheme, by performing feature processing on user behavior data, a user behavior data framework is constructed, and based on the user behavior data framework, the user's search information is predicted, thereby improving the accuracy of identifying the user's future search intent.
[0078] Optionally, a user behavior data framework is constructed based on each behavioral feature data, including: inputting each behavioral feature data into a data framework model to obtain an initial user behavior data framework; and performing data association processing on each behavioral feature data within the initial user behavior data framework to obtain the final user behavior data framework.
[0079] In this embodiment, the terminal inputs each behavioral feature data into a data framework model to obtain an initial user behavior data framework. Within this initial framework, the terminal determines the relationships between each behavioral feature data point and performs data association processing between related behavioral feature data points to obtain the final user behavior data framework. This user behavior data framework contains all the user's behavioral feature data and the relationships between them.
[0080] Based on the above scheme, a user behavior data framework is established to provide a data foundation for the subsequent intent recognition model to identify the user's search intent.
[0081] Optionally, the intent recognition model includes neural networks, update gates, and reset gates; correspondingly, such as Figure 2 As shown, the intent recognition model, based on the user behavior data framework, predicts and identifies the user's search intent, obtaining the user's predicted search information, including:
[0082] Step S201: Obtain the user's current search information.
[0083] Step S202: Through a neural network, the search intent of the user behavior data framework is identified, and among the current search information, the current search information that is similar to the search intent is selected as the first predicted search information.
[0084] Step S203: Through the update gate, select current search information similar to the search intent from the current search information other than the first predicted search information, and use it as the second predicted search information.
[0085] Step S204: By resetting the gate, calculate the correlation value of the first predicted search information and the second predicted search information, and if the correlation value is greater than the preset correlation threshold, merge the first predicted search information and the second predicted search information to obtain the initial predicted search information.
[0086] Step S205: If the preset iteration termination condition is not met, return to the step of identifying the user's search intent through the neural network and selecting the current search information that is similar to the search intent from each current search information as the first predicted search information step, until the preset iteration termination condition is met, and then use all the initial predicted search information as the user's predicted search information.
[0087] In this embodiment, the intent recognition model is a deep neural network based on gated recurrent units. This intent recognition model includes a neural network, an update gate, and a reset gate, such as... Figure 3 As shown, the input layer, embedding layer, feedforward layer, and output layer are all deep neural networks, while the GRU (Gated Recurrent Unit) layer contains update gates and reset gates. The terminal pre-acquires information that the user has not yet searched (i.e., current search information), and first uses a neural network to perform preliminary search intent recognition processing on the user behavior data framework to obtain the search intent related to the user. Based on this search intent, the terminal selects current search information similar to the current search intent from among the current search information as the first predicted search information. Here, the search intent is the intention information of the user to perform a search operation. For example, the search intent could be to query related content in a certain window through voice search. The terminal filters each current search information item by item based on the above information and selects current search information similar to the search intent as the first predicted search information. This first predicted search information can be, but is not limited to, one or more items.
[0088] The terminal uses an update gate to select, from the current search information (excluding the first predicted search information), current search information similar to the search intent, as the second predicted search information. Using a reset gate, the terminal calculates the correlation value between the first and second predicted search information. If the correlation value is greater than a preset correlation threshold, the terminal merges the first and second predicted search information to obtain the initial predicted search information.
[0089] The specific execution methods for updating and resetting the door are as follows:
[0090] The gated loop unit calculates the hidden state (i.e., selects the current search information that is similar to the search intent) by introducing the concepts of reset gate and update gate.
[0091] The calculation method for updating and resetting doors is shown in the following formula:
[0092] R t =σ(X) t W xr +H t-1 W hr +b r )
[0093] Z t =σ(X) t W xr +H t-1 W hr +b Z )
[0094] In the above formula, R t Indicates the update gate, Z t This represents the reset gate, where σ is a constant parameter and b r To update the gate parameters, b Z To reset the gate parameters, H t-1 X represents the state information from the previous time step (i.e., the prediction search information from the previous prediction). t For search intent, W xr With W hr This refers to the different behavioral characteristics of users within the user behavior data framework.
[0095] The gated recurrent unit then assists in the calculation of the hidden state (i.e., the initial predicted search information) by calculating the candidate activation.
[0096] H′ t =tanh(X) t W xh +(R t ⊙H t-1 W hh +b h )
[0097] H t =Z t ⊙H t-1 +(1-Z t )⊙H′ t
[0098] In the above formula, H′ t Z represents the candidate hidden state. t Indicates resetting the door, R t Indicates the update gate, b h H represents the parameters of the candidate hidden state. t In the hidden state, X t For search intent, Wxh With W hh This refers to the different behavioral characteristics of users within the user behavior data framework.
[0099] The reset gate controls how the previous hidden state flows into the current candidate hidden state; the update gate controls how the candidate hidden state containing the current time step information updates the current hidden state.
[0100] The terminal presets an iteration termination condition and, after obtaining the initial predicted search information, determines whether the current iteration meets the iteration termination condition. If the iteration termination condition is not met, the terminal returns to step S202. This process continues until the iteration termination condition is met, at which point the terminal uses the initial predicted search information obtained from the last iteration as the user's predicted search information. The preset iteration termination condition may include, but is not limited to, the number of iterations, the number of predicted search information items, etc.
[0101] Specifically, such as Figure 4 As shown, the terminal acquires multiple behavioral data points for both cold-start and non-cold-start users through the aforementioned steps. For each user, the terminal performs feature adjustment processing on each behavioral data point to obtain multiple behavioral feature data points for that user. Based on these behavioral feature data points, the terminal constructs a user behavior data framework for each user. For cold-start users, the terminal uses the intent recognition model (GRU sequence model 1) corresponding to the cold-start user and, based on the cold-start user behavior data framework, identifies the user's search intent to obtain the cold-start user's search intent. It then predicts the cold-start user's search intent to obtain the predicted search information (i.e., GRU intent prediction result features). For non-cold-start users, the terminal uses the intent recognition model (GRU sequence model 2) corresponding to the non-cold-start user and, based on the non-cold-start user behavior data framework, predicts the non-cold-start user's search information to obtain the predicted search information (i.e., GRU intent prediction result features).
[0102] Based on the above scheme, a deep neural network based on gated recurrent units is used to predict users' search information, thereby improving the accuracy of identifying users' future search intentions.
[0103] Optionally, the method further includes: acquiring the sample behavioral feature data of each sample user and the sample search information of the sample user; inputting the sample behavioral feature data of each sample user and the sample search information of the sample user into the initial intent recognition model, training the initial intent recognition model, and obtaining the intent recognition model.
[0104] In this embodiment, before performing intent recognition on the user, the terminal first acquires the sample user's behavioral feature data and the sample user's search information. The terminal inputs the sample user's behavioral feature data and the sample user's search information into an initial intent recognition model and trains the initial intent recognition model to obtain an intent recognition model. The number of sample users is not limited to one; the training process can be, but is not limited to, a single training iteration. Alternatively, the intent recognition model can be iteratively trained using data from multiple sample users to obtain the final model. The sample users are non-cold-start users.
[0105] Based on the above scheme, by training the initial intent recognition model, an intent recognition model is obtained, which improves the accuracy of the intent recognition model in recognizing user intent.
[0106] Optionally, after obtaining the intent recognition model, the process further includes: acquiring the behavioral feature data of offline users and their search information; determining the predicted search information of offline users based on the behavioral feature data using the intent recognition model; and returning to the step of acquiring the sample behavioral feature data of sample users and their sample search information if the similarity between the predicted search information of offline users and their search information is less than a preset similarity value, until the similarity between the predicted search information of offline users and their search information is not less than a preset similarity value, and then outputting the intent recognition model obtained in the last iteration.
[0107] In this embodiment, with the authorization of the offline user, the terminal obtains the offline user's behavioral feature data and search information from the database. Using a trained intent recognition model, the terminal simulates and predicts the user's search information based on the behavioral feature data, obtaining the predicted search information for the offline user. The terminal then calculates the similarity between the predicted search information and the actual offline user's search information using a similarity algorithm, and determines whether the similarity is less than a preset similarity value.
[0108] If the similarity between the predicted search information and the offline user's search information is not less than a preset similarity value, the terminal outputs the intent recognition model. If the similarity between the predicted search information and the offline user's search information is less than the preset similarity value, the terminal returns to the previous step and iterates the above operation until the predicted search information and the offline user's search information are not less than the preset similarity value. Then, the terminal outputs the intent recognition model obtained from the final iteration. Here, the offline user is the cold-start user.
[0109] Specifically, the model training process is mainly implemented in two stages. In the first stage, GRU model 1 is initialized and trained using offline batches of full user data to serve cold-start users. In the second stage, based on the initialized and deployed model 1, transfer learning is performed on the mobile terminal's machine learning framework. The training data uses the single user data from this mobile terminal to obtain model 2 for non-cold-start users logged in on the current device. The model invocation logic priority in the model application stage is model 2 > model 1.
[0110] Preparation of model training data, such as Figure 5 As shown, the system first classifies the detailed behavioral data generated by users on mobile devices into scene classification sequences based on behavioral data classification. The data is segmented using a fixed-length batch method. The segmentation of the training set is different from the proportional segmentation used by other algorithms. It needs to be segmented according to the time sequence logic.
[0111] The following is an example of a fixed-length batch segmentation sequence:
[0112] User ID Original sequence Input sequence (X) Intent Classification (Y) A C1, C7, C2, C3, C9 {C1,C7},{C7,C2},{C2,C3} {C2}, {C3}, {C9} B C9,C6,C4 {C9,C6} {C6},{C4} C C22, C13, C1, C2 {C22, C13}, {C13, C1} {C1},{C2} ... ... ... ...
[0113] The dataset constructed from the segmented and processed sequence X and the target intent classification Y can be used for model training and similarity calculation for predicting search information.
[0114] The formula for calculating the similarity of predicted search information is as follows:
[0115]
[0116] Where, r′ s ,ii represents positive samples of search information from offline users, r′ s ,jj represents negative samples of search information from offline users, σ is a constant parameter, N s F is the similarity value used to predict search information.
[0117] Based on the above scheme, the trained intent recognition model is tested, and if the trained intent recognition model meets the detection conditions, the trained intent recognition model is output. This improves the accuracy of the intent recognition model in recognizing user search intent.
[0118] Optionally, after obtaining the user's predicted search information, the method further includes: acquiring the user's current behavioral feature data and the user's search information; inputting the current behavioral feature data and the user's search information into the intent recognition model, adjusting the intent recognition model, and obtaining the adjusted intent recognition model.
[0119] In this embodiment, the terminal reacquires the user's current behavioral feature data and the user's search information. The terminal inputs the user's current behavioral feature data and the user's search information into the intent recognition model, and trains and optimizes the intent recognition model (adjusting the intent recognition model) to obtain an optimized intent recognition model (obtaining an adjusted intent recognition model).
[0120] Based on the above scheme, by further optimizing the intent recognition model after each user's search intent is identified, the accuracy of the intent recognition model in identifying the user's search intent is improved.
[0121] This application also provides an example of search intent recognition, such as Figure 6 As shown, the specific processing procedure includes the following steps:
[0122] Step S601: Obtain multiple behavioral data of the user.
[0123] Step S602: Perform feature adjustment processing on each behavioral data to obtain multiple behavioral feature data.
[0124] Step S603: Input the behavioral feature data into the data framework model to obtain the initial user behavior data framework.
[0125] Step S604: In the initial user behavior data framework, perform data association processing on each behavior feature data to obtain the user behavior data framework.
[0126] Step S605: Obtain the user's current search information.
[0127] Step S606: Through a neural network, identify the search intent of the user behavior data framework, and select the current search information that is similar to the search intent from among the current search information as the first predicted search information.
[0128] Step S607: Through the update gate, select current search information similar to the search intent from the current search information other than the first predicted search information, and use it as the second predicted search information.
[0129] Step S608: By resetting the gate, calculate the correlation value of the first predicted search information and the second predicted search information, and if the correlation value is greater than a preset correlation threshold, merge the first predicted search information and the second predicted search information to obtain the initial predicted search information.
[0130] Step S609: Determine whether the preset iteration termination condition is met.
[0131] If yes, use all initial predicted search information as the user's predicted search information and proceed to step S610; otherwise, proceed to step S606.
[0132] Step S610: Using the intent recognition model and based on the user behavior data framework, predict the user's search intent to obtain the user's predicted search information.
[0133] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.
[0134] Based on the same inventive concept, this application also provides a search intent recognition device for implementing the search intent recognition method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more search intent recognition device embodiments provided below can be found in the limitations of the search intent recognition method described above, and will not be repeated here.
[0135] In one embodiment, such as Figure 7 As shown, a search intent recognition device is provided, including: an acquisition module 710, a processing module 720, a construction module 730, and a prediction module 740, wherein:
[0136] Module 710 is used to acquire multiple behavioral data of the user;
[0137] Processing module 720 is used to perform feature adjustment processing on each behavioral data to obtain multiple behavioral feature data; the behavioral feature data is the characteristic data information of the behavioral data.
[0138] Module 730 is used to build a user behavior data framework based on various behavioral feature data. The user behavior data framework is a data framework that includes all user behavioral feature data and the relationships between various behavioral feature data.
[0139] The prediction module 740 is used to predict the user's search intent based on the user behavior data framework through the intent recognition model, and obtain the user's predicted search information.
[0140] Optional, build module 730, specifically used for:
[0141] Input the behavioral feature data into the data framework model to obtain the initial user behavior data framework;
[0142] In the initial user behavior data framework, data association processing is performed on each behavioral feature data to obtain the user behavior data framework.
[0143] Optionally, the intent recognition model includes a neural network, an update gate, and a reset gate; the prediction module 740 is specifically used for:
[0144] Obtain the user's current search information;
[0145] The search intent of the user behavior data framework is identified through neural networks, and the current search information that is similar to the search intent is selected from the current search information as the first predicted search information.
[0146] By updating the gate, among the current search information other than the first predicted search information, the current search information that is similar to the search intent is selected as the second predicted search information;
[0147] By resetting the gate, the correlation value of the first predicted search information and the second predicted search information is calculated. If the correlation value is greater than the preset correlation threshold, the first predicted search information and the second predicted search information are merged to obtain the initial predicted search information.
[0148] If the preset iteration termination condition is not met, return to the search intent identified by the neural network in the user behavior data framework, and select the current search information similar to the search intent from the current search information as the first predicted search information step, until the preset iteration termination condition is met, and then use all the initial predicted search information as the user's predicted search information.
[0149] Optionally, the device also includes:
[0150] The sample acquisition module is used to acquire the sample user's behavioral feature data and sample user's search information;
[0151] The training module is used to input the sample user's behavioral feature data and sample user's search information into the initial intent recognition model, train the initial intent recognition model, and obtain the intent recognition model.
[0152] Optionally, the device also includes:
[0153] The feature acquisition module is used to acquire various behavioral feature data of offline users, as well as their search information;
[0154] The offline prediction module is used to determine the predicted search information of offline users based on various behavioral feature data through an intent recognition model.
[0155] The optimization module is used to return the steps of obtaining the sample user's behavioral feature data and sample user's search information when the similarity between the predicted search information of the offline user and the offline user's search information is less than a preset similarity value. This process continues until the similarity between the predicted search information of the offline user and the offline user's search information is not less than a preset similarity value, at which point the intent recognition model obtained from the last iteration is output.
[0156] Optionally, the device also includes:
[0157] The re-acquisition module is used to acquire the user's current behavioral characteristics data, as well as the user's search information;
[0158] The adjustment module is used to input the current behavioral feature data and the user's search information into the intent recognition model, adjust the intent recognition model, and obtain the adjusted intent recognition model.
[0159] Each module in the aforementioned search intent recognition device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.
[0160] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 8 As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a search intent recognition method. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.
[0161] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0162] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the following steps:
[0163] Obtain multiple behavioral data points from users;
[0164] Each behavioral data point is subjected to feature adjustment processing to obtain multiple behavioral feature data; the behavioral feature data is the characteristic data information of the behavioral data.
[0165] Based on the aforementioned behavioral feature data, a user behavior data framework is constructed; the user behavior data framework is a data framework that includes all behavioral feature data of the user, as well as the correlation relationships between the aforementioned behavioral feature data.
[0166] By using the intent recognition model and based on the user behavior data framework, the user's search intent is predicted to obtain the user's predicted search information.
[0167] Optionally, the step of constructing a user behavior data framework based on each of the behavioral feature data includes:
[0168] Input the behavioral feature data into the data framework model to obtain the initial user behavior data framework;
[0169] In the initial user behavior data framework, data association processing is performed on each of the behavioral feature data to obtain the user behavior data framework.
[0170] Optionally, the intent recognition model includes a neural network, an update gate, and a reset gate; the step of predicting and recognizing the user's search intent based on the user behavior data framework using the intent recognition model to obtain the user's predicted search information includes:
[0171] Obtain the user's current search information;
[0172] The neural network identifies the search intent of the user behavior data framework, and selects the current search information that is similar to the search intent from among the current search information as the first predicted search information.
[0173] Through the update gate, among the current search information other than each of the first predicted search information, the current search information that is similar to the search intent is selected as the second predicted search information;
[0174] The first predicted search information and the second predicted search information are calculated through the reset gate. If the correlation value is greater than a preset correlation threshold, the first predicted search information and the second predicted search information are merged to obtain the initial predicted search information.
[0175] If the preset iteration termination condition is not met, the process returns to identifying the search intent of the user behavior data framework through the neural network, and selecting current search information similar to the search intent from each current search information as the first predicted search information step, until the preset iteration termination condition is met, and then all the initial predicted search information is used as the user's predicted search information.
[0176] Optionally, the method further includes:
[0177] Obtain the behavioral feature data of each sample user, as well as the sample search information of the sample users;
[0178] The sample user's behavioral feature data and the sample user's search information are input into the initial intent recognition model, and the initial intent recognition model is trained to obtain the intent recognition model.
[0179] Optionally, after obtaining the intent recognition model, the method further includes:
[0180] Acquire the behavioral characteristic data of offline users, as well as the search information of the offline users;
[0181] Based on the behavioral feature data, the predicted search information of the offline user is determined using an intent recognition model.
[0182] If the similarity between the predicted search information of the offline user and the search information of the offline user is less than a preset similarity value, return to the step of obtaining the sample behavioral feature data of each sample user and the sample search information of the sample user, until the similarity between the predicted search information of the offline user and the search information of the offline user is not less than the preset similarity value, and output the intent recognition model obtained in the last iteration.
[0183] Optionally, after obtaining the user's predicted search information, the method further includes:
[0184] Obtain the user's current behavioral characteristics data and the user's search information;
[0185] The current behavioral feature data and the user's search information are input into the intent recognition model, and the intent recognition model is adjusted to obtain the adjusted intent recognition model.
[0186] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, performs the following steps:
[0187] Obtain multiple behavioral data points from users;
[0188] Each behavioral data point is subjected to feature adjustment processing to obtain multiple behavioral feature data; the behavioral feature data is the characteristic data information of the behavioral data.
[0189] Based on the aforementioned behavioral feature data, a user behavior data framework is constructed; the user behavior data framework is a data framework that includes all behavioral feature data of the user, as well as the correlation relationships between the aforementioned behavioral feature data.
[0190] By using the intent recognition model and based on the user behavior data framework, the user's search intent is predicted to obtain the user's predicted search information.
[0191] Optionally, the step of constructing a user behavior data framework based on each of the behavioral feature data includes:
[0192] Input the behavioral feature data into the data framework model to obtain the initial user behavior data framework;
[0193] In the initial user behavior data framework, data association processing is performed on each of the behavioral feature data to obtain the user behavior data framework.
[0194] Optionally, the intent recognition model includes a neural network, an update gate, and a reset gate; the step of predicting and recognizing the user's search intent based on the user behavior data framework using the intent recognition model to obtain the user's predicted search information includes:
[0195] Obtain the user's current search information;
[0196] The neural network identifies the search intent of the user behavior data framework, and selects the current search information that is similar to the search intent from among the current search information as the first predicted search information.
[0197] Through the update gate, among the current search information other than each of the first predicted search information, the current search information that is similar to the search intent is selected as the second predicted search information;
[0198] The first predicted search information and the second predicted search information are calculated through the reset gate. If the correlation value is greater than a preset correlation threshold, the first predicted search information and the second predicted search information are merged to obtain the initial predicted search information.
[0199] If the preset iteration termination condition is not met, the process returns to identifying the search intent of the user behavior data framework through the neural network, and selecting current search information similar to the search intent from each current search information as the first predicted search information step, until the preset iteration termination condition is met, and then all the initial predicted search information is used as the user's predicted search information.
[0200] Optionally, the method further includes:
[0201] Obtain the behavioral feature data of each sample user, as well as the sample search information of the sample users;
[0202] The sample user's behavioral feature data and the sample user's search information are input into the initial intent recognition model, and the initial intent recognition model is trained to obtain the intent recognition model.
[0203] Optionally, after obtaining the intent recognition model, the method further includes:
[0204] Acquire the behavioral characteristic data of offline users, as well as the search information of the offline users;
[0205] Based on the behavioral feature data, the predicted search information of the offline user is determined using an intent recognition model.
[0206] If the similarity between the predicted search information of the offline user and the search information of the offline user is less than a preset similarity value, return to the step of obtaining the sample behavioral feature data of each sample user and the sample search information of the sample user, until the similarity between the predicted search information of the offline user and the search information of the offline user is not less than the preset similarity value, and output the intent recognition model obtained in the last iteration.
[0207] Optionally, after obtaining the user's predicted search information, the method further includes:
[0208] Obtain the user's current behavioral characteristics data and the user's search information;
[0209] The current behavioral feature data and the user's search information are input into the intent recognition model, and the intent recognition model is adjusted to obtain the adjusted intent recognition model.
[0210] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:
[0211] Obtain multiple behavioral data points from users;
[0212] Each behavioral data point is subjected to feature adjustment processing to obtain multiple behavioral feature data; the behavioral feature data is the characteristic data information of the behavioral data.
[0213] Based on the aforementioned behavioral feature data, a user behavior data framework is constructed; the user behavior data framework is a data framework that includes all behavioral feature data of the user, as well as the correlation relationships between the aforementioned behavioral feature data.
[0214] By using the intent recognition model and based on the user behavior data framework, the user's search intent is predicted to obtain the user's predicted search information.
[0215] Optionally, the step of constructing a user behavior data framework based on each of the behavioral feature data includes:
[0216] Input the behavioral feature data into the data framework model to obtain the initial user behavior data framework;
[0217] In the initial user behavior data framework, data association processing is performed on each of the behavioral feature data to obtain the user behavior data framework.
[0218] Optionally, the intent recognition model includes a neural network, an update gate, and a reset gate; the step of predicting and recognizing the user's search intent based on the user behavior data framework using the intent recognition model to obtain the user's predicted search information includes:
[0219] Obtain the user's current search information;
[0220] The neural network identifies the search intent of the user behavior data framework, and selects the current search information that is similar to the search intent from among the current search information as the first predicted search information.
[0221] Through the update gate, among the current search information other than each of the first predicted search information, the current search information that is similar to the search intent is selected as the second predicted search information;
[0222] The first predicted search information and the second predicted search information are calculated through the reset gate. If the correlation value is greater than a preset correlation threshold, the first predicted search information and the second predicted search information are merged to obtain the initial predicted search information.
[0223] If the preset iteration termination condition is not met, the process returns to identifying the search intent of the user behavior data framework through the neural network, and selecting current search information similar to the search intent from each current search information as the first predicted search information step, until the preset iteration termination condition is met, and then all the initial predicted search information is used as the user's predicted search information.
[0224] Optionally, the method further includes:
[0225] Obtain the behavioral feature data of each sample user, as well as the sample search information of the sample users;
[0226] The sample user's behavioral feature data and the sample user's search information are input into the initial intent recognition model, and the initial intent recognition model is trained to obtain the intent recognition model.
[0227] Optionally, after obtaining the intent recognition model, the method further includes:
[0228] Acquire the behavioral characteristic data of offline users, as well as the search information of the offline users;
[0229] Based on the behavioral feature data, the predicted search information of the offline user is determined using an intent recognition model.
[0230] If the similarity between the predicted search information of the offline user and the search information of the offline user is less than a preset similarity value, return to the step of obtaining the sample behavioral feature data of each sample user and the sample search information of the sample user, until the similarity between the predicted search information of the offline user and the search information of the offline user is not less than the preset similarity value, and output the intent recognition model obtained in the last iteration.
[0231] Optionally, after obtaining the user's predicted search information, the method further includes:
[0232] Obtain the user's current behavioral characteristics data and the user's search information;
[0233] The current behavioral feature data and the user's search information are input into the intent recognition model, and the intent recognition model is adjusted to obtain the adjusted intent recognition model.
[0234] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.
[0235] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.
[0236] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0237] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.
Claims
1. A method for identifying search intent, characterized in that, The method includes: Obtain multiple behavioral data points from users; Each behavioral data point is subjected to feature adjustment processing to obtain multiple behavioral feature data; the behavioral feature data is the characteristic data information of the behavioral data. Each of the aforementioned behavioral feature data is input into the data framework model to obtain an initial user behavior data framework; in the initial user behavior data framework, data association processing is performed on each of the aforementioned behavioral feature data to obtain a user behavior data framework; the user behavior data framework is a data framework that includes all the user's behavioral feature data and the association relationships between each of the aforementioned behavioral feature data, and the intent recognition model includes a neural network, an update gate, and a reset gate. Obtain the user's current search information; The neural network identifies the search intent of the user behavior data framework, and selects the current search information that is similar to the search intent from among the current search information as the first predicted search information. Through the update gate, among the current search information other than each of the first predicted search information, the current search information that is similar to the search intent is selected as the second predicted search information; The first predicted search information and the second predicted search information are calculated through the reset gate. If the correlation value is greater than a preset correlation threshold, the first predicted search information and the second predicted search information are merged to obtain the initial predicted search information. If the preset iteration termination condition is not met, the process returns to identifying the search intent of the user behavior data framework through the neural network, and selecting current search information similar to the search intent from each current search information as the first predicted search information step, until the preset iteration termination condition is met, and then all the initial predicted search information is used as the user's predicted search information.
2. The method according to claim 1, characterized in that, The method further includes: Obtain the behavioral feature data of each sample user, as well as the sample search information of the sample users; The sample user's behavioral feature data and the sample user's search information are input into the initial intent recognition model, and the initial intent recognition model is trained to obtain the intent recognition model.
3. The method according to claim 2, characterized in that, After obtaining the intent recognition model, the process further includes: Acquire the behavioral characteristic data of offline users, as well as the search information of the offline users; Based on the behavioral feature data, the predicted search information of the offline user is determined using an intent recognition model. If the similarity between the predicted search information of the offline user and the search information of the offline user is less than a preset similarity value, return to the step of obtaining the sample behavioral feature data of each sample user and the sample search information of the sample user, until the similarity between the predicted search information of the offline user and the search information of the offline user is not less than the preset similarity value, and output the intent recognition model obtained in the last iteration.
4. The method according to claim 1, characterized in that, After obtaining the user's predicted search information, the method further includes: Obtain the user's current behavioral characteristics data and the user's search information; The current behavioral feature data and the user's search information are input into the intent recognition model, and the intent recognition model is adjusted to obtain the adjusted intent recognition model.
5. A search intent recognition device, characterized in that, The device includes: The acquisition module is used to acquire multiple behavioral data of the user; The processing module is used to perform feature adjustment processing on each of the behavioral data to obtain multiple behavioral feature data; the behavioral feature data is the characteristic data information of the behavioral data; A construction module is used to input the behavioral feature data into a data framework model to obtain an initial user behavior data framework; in the initial user behavior data framework, data association processing is performed on the behavioral feature data to obtain a user behavior data framework; the user behavior data framework is a data framework that includes all behavioral feature data of the user and the association relationships between the behavioral feature data, and the intent recognition model includes a neural network, an update gate, and a reset gate. The prediction module is used to obtain the user's current search information; identify the search intent of the user behavior data framework through the neural network, and select current search information similar to the search intent from each current search information as the first predicted search information; select current search information similar to the search intent from the current search information other than the first predicted search information through the update gate as the second predicted search information; calculate the correlation value of the first predicted search information and the second predicted search information through the reset gate, and merge the first predicted search information and the second predicted search information if the correlation value is greater than a preset correlation threshold to obtain initial predicted search information; if the preset iteration termination condition is not met, return to the step of identifying the search intent of the user behavior data framework through the neural network and selecting current search information similar to the search intent from each current search information as the first predicted search information, until the preset iteration termination condition is met, and then use all the initial predicted search information as the user's predicted search information.
6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 4.
7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.
8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.