Location-based search processing method, apparatus, device, and storage medium

Abstract

The application provides a location-based search processing method, device and equipment and a storage medium. The method comprises the following steps: in response to a location search request of a user, inputting search information into a preset multi-task search model; simultaneously processing the search information based on a city jump decision task and a search suggestion task in the multi-task search model to obtain target city location features corresponding to the city jump decision task and search suggestion features corresponding to the search suggestion task; obtaining a search processing result based on the target city location features and the search suggestion features; obtaining a search suggestion list with the highest matching degree with the search processing result in a preset search suggestion library, and displaying the search suggestion list on a search interface. Through the above method, the training and development and maintenance costs of multiple models can be effectively reduced.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a location-based search processing method, apparatus, device, and storage medium. Background Technology

[0002] With the increasing popularity of electronic maps, users are demanding more and more intelligent location search functions from them.

[0003] In current location search scenarios, map software's cross-city location search and suggested search functions provide users with greater convenience. For example, when a user searches for a location, the map software can intelligently jump to the corresponding city location or recommend more suggested search locations. Related technologies utilize two separate network models within the map software to sequentially execute the city location jump task and the location search suggestion task, which results in high development and training costs for the models. Summary of the Invention

[0004] This application provides a location-based search processing method, apparatus, device, and storage medium to address the problem of high model development and training costs.

[0005] According to one aspect of this application, a location-based search processing method is provided, comprising:

[0006] In response to a user's location search request, search information is input into a pre-set multi-task search model; wherein, the multi-task search model is a model obtained by jointly training multiple tasks, including a city jump decision task and a search suggestion task;

[0007] The search information is processed simultaneously using the city jump decision task and the search suggestion task in the multi-task search model to obtain the target city location features corresponding to the city jump decision task and the search suggestion features corresponding to the search suggestion task; and the search processing results are obtained based on the target city location features and the search suggestion features.

[0008] From the preset search suggestion library, obtain the list of search suggestions that best match the search results, and display the list of search suggestions on the search interface.

[0009] In one implementation, the method of training the multi-task search model includes:

[0010] Obtain a training dataset, which includes a sample set of city jump decision tasks and a sample set of search suggestion tasks;

[0011] Sample features are extracted based on the training dataset. These sample features include common features, city analysis-related input features, and search suggestion-related input features.

[0012] The initial neural network is trained based on the sample features to obtain the multi-task search model; wherein, the initial neural network includes: an initial city jump decision task and an initial search suggestion task.

[0013] In one implementation, obtaining the training dataset includes:

[0014] Obtain historical behavior log data for at least one user, and obtain several data samples based on the historical behavior log data;

[0015] Each data sample is classified according to its task sample type, and the city jump decision task sample set and search suggestion task sample set are obtained based on the classification results to obtain the training dataset.

[0016] In one implementation, the training dataset carries identification information for each data sample, the identification information including at least one of the following: user identifier, terminal identifier, and log generation time;

[0017] The extraction of sample features based on the training dataset includes:

[0018] For each sample in the training dataset, obtain the service features corresponding to the identification information of the sample from the feature service log data;

[0019] According to the preset feature classification reference table, the service features are classified to obtain sample features including public features, city analysis-related input features, and search suggestion-related input feature types.

[0020] In one embodiment, the initial neural network includes: an input layer; a data sharing layer connected to the input layer; a city analysis feature extraction layer and a search suggestion feature extraction layer connected to the data sharing layer, respectively; and an output layer connected to the city analysis feature extraction layer and the search suggestion feature extraction layer, respectively.

[0021] The process of training an initial neural network based on the sample features to obtain the multi-task search model includes:

[0022] The sample features are input into the input layer;

[0023] If the sample feature is a common feature, the input layer is triggered to transmit the common feature to the data sharing layer; and / or, if the sample feature is a city analysis-related input feature, the input layer is triggered to transmit the city analysis-related input feature to the city analysis feature extraction layer; and / or, if the sample feature is a search suggestion-related input feature, the input layer is triggered to transmit the search suggestion-related input feature to the search suggestion feature extraction layer.

[0024] The data sharing layer is triggered to simultaneously train the initial city jump decision task and the initial search suggestion task based on the common features, and the training results are sent to the city analysis feature extraction layer and the search suggestion feature extraction layer, respectively.

[0025] The city analysis feature extraction layer is triggered to obtain the target city location features based on the training processing results and the city analysis-related input features; and the search suggestion feature extraction layer is triggered to obtain the search suggestion features based on the training processing results and the search suggestion-related input features.

[0026] The output layer is triggered to output the trained target city location features and the trained search suggestion features; and the loss value between the trained target city location features and the trained search suggestion features and the preset target features is obtained;

[0027] The first weight corresponding to the initial city jump decision task and the second weight corresponding to the initial search suggestion task are adjusted according to the loss value to obtain the trained city jump decision task and the trained search suggestion task.

[0028] Based on the trained city jump decision task and the trained search suggestion task, a trained deep neural network is obtained.

[0029] In one implementation, it further includes:

[0030] Determine the training parameters in the initial neural network, wherein the training parameters include the learning rate and / or the processing batch;

[0031] The step of adjusting the first weight corresponding to the initial city jump decision task and adjusting the second weight corresponding to the initial search suggestion task based on the loss value includes:

[0032] Based on the training parameters, the first weight corresponding to the initial city jump decision task and the second weight corresponding to the initial search suggestion task are adjusted according to the loss value.

[0033] In one embodiment, the method further includes:

[0034] The preset multi-task search model is periodically updated according to a preset cycle.

[0035] The periodic update process includes:

[0036] The log data corresponding to the search processing performed by the multi-task model is written to the search log database in real time, so as to obtain a new training dataset for update processing based on the log database.

[0037] Obtain the duration of the search processing performed by the multi-task model;

[0038] When the duration reaches the calibrated value, the multi-task model is trained using the new training dataset, and the trained multi-task model is used as the updated multi-task training model.

[0039] According to another aspect of this application, a location-based search processing apparatus is provided, comprising:

[0040] The input module is configured to input search information into a pre-set multi-task search model in response to a user's location search request; wherein the multi-task search model is a model obtained by jointly training multiple tasks, including a city jump decision task and a search suggestion task.

[0041] The model processing module is configured to simultaneously process the search information based on the city jump decision task and the search suggestion task in the multi-task search model, so as to obtain the target city location features corresponding to the city jump decision task and the search suggestion features corresponding to the search suggestion task; and obtain the search processing results based on the target city location features and the search suggestion features.

[0042] The matching module is configured to retrieve a list of search suggestions that have the highest matching degree with the search processing result from a preset search suggestion library, and display the list of search suggestions on the search interface.

[0043] In one implementation, the module for training the multi-task search model includes:

[0044] The dataset acquisition module is configured to acquire a training dataset, which includes a city jump decision task sample set and a search suggestion task sample set.

[0045] The feature extraction module is configured to extract sample features based on the training dataset, the sample features including common features, city analysis-related input features, and search suggestion-related input features;

[0046] The training module is configured to train an initial neural network based on the sample features to obtain the multi-task search model; wherein the initial neural network includes an initial city jump decision task and an initial search suggestion task.

[0047] In one implementation, the dataset acquisition module includes:

[0048] The log acquisition unit is configured to acquire historical behavior log data corresponding to at least one user, and acquire several data samples based on the historical behavior log data.

[0049] The classification unit is configured to classify each data sample according to the task sample type, and obtain the city jump decision task sample set and the search suggestion task sample set based on the classification results to obtain the training dataset.

[0050] In one implementation, the training dataset carries identification information for each data sample, the identification information including at least one of the following: user identifier, terminal identifier, and log generation time;

[0051] The feature extraction module includes:

[0052] The service feature acquisition unit is configured to acquire service features corresponding to the identification information of each sample from the feature service log data for each sample in the training dataset.

[0053] The feature classification unit is configured to classify the service features according to a preset feature classification reference table to obtain sample features including public features, city analysis-related input features, and search suggestion-related input feature types.

[0054] In one embodiment, the initial neural network includes: an input layer; a data sharing layer connected to the input layer; a city analysis feature extraction layer and a search suggestion feature extraction layer connected to the data sharing layer, respectively; and an output layer connected to the city analysis feature extraction layer and the search suggestion feature extraction layer, respectively.

[0055] The training module includes:

[0056] An input unit is configured to input the sample features into the input layer;

[0057] The first triggering unit is configured to, if the sample feature is a common feature, trigger the input layer to transmit the common feature to the data sharing layer; and / or, if the sample feature is a city analysis-related input feature, trigger the input layer to transmit the city analysis-related input feature to the city analysis feature extraction layer; and / or, if the sample feature is a search suggestion-related input feature, trigger the input layer to transmit the search suggestion-related input feature to the search suggestion feature extraction layer.

[0058] The second triggering unit is configured to trigger the data sharing layer to simultaneously train the initial city jump decision task and the initial search suggestion task based on the public features, and send the training results to the city analysis feature extraction layer and the search suggestion feature extraction layer respectively.

[0059] The third triggering unit is configured to trigger the city analysis feature extraction layer to obtain the target city location features based on the training processing results and the city analysis-related input features; and to trigger the search suggestion feature extraction layer to obtain the search suggestion features based on the training processing results and the search suggestion-related input features.

[0060] The fourth triggering unit is configured to trigger the output layer to output the trained target city location features and the trained search suggestion features; and to obtain the loss value between the trained target city location features and the trained search suggestion features and the preset target features;

[0061] The weight adjustment unit is configured to adjust the first weight corresponding to the initial city jump decision task and the second weight corresponding to the initial search suggestion task according to the loss value, so as to obtain the trained city jump decision task and the trained search suggestion task.

[0062] The acquisition unit is configured to obtain a trained deep neural network based on the trained city jump decision task and the trained search suggestion task.

[0063] In one implementation, it further includes:

[0064] A parameter determination module is configured to determine training parameters in the initial neural network, wherein the training parameters include a learning rate and / or a batch size.

[0065] The weight adjustment unit is specifically configured to adjust the first weight corresponding to the initial city jump decision task and the second weight corresponding to the initial search suggestion task based on the training parameters and the loss value.

[0066] In one embodiment, the device further includes:

[0067] The update module is configured to periodically update the preset multi-task search model according to a preset cycle.

[0068] The periodic update process includes:

[0069] The log data corresponding to the search processing performed by the multi-task model is written to the search log database in real time, so as to obtain a new training dataset for update processing based on the log database.

[0070] Obtain the duration of the search processing performed by the multi-task model;

[0071] When the duration reaches the calibrated value, the multi-task model is trained using the new training dataset, and the trained multi-task model is used as the updated multi-task training model.

[0072] According to another aspect of this application, an electronic device is provided, comprising: a memory and a processor;

[0073] The memory stores computer-executed instructions;

[0074] The processor executes computer execution instructions stored in the memory, causing the electronic device to perform the location-based search processing method.

[0075] According to another aspect of this application, a computer-readable storage medium is provided, wherein computer-executable instructions are stored therein, which, when executed by a processor, are used to implement the location-based search processing method described above.

[0076] The location-based search processing method, apparatus, device, and storage medium provided in this application, in response to a user's location search request, inputs search information into a pre-set multi-task search model; simultaneously processes the search information based on a city jump decision task and a search suggestion task within the multi-task search model to obtain target city location features corresponding to the city jump decision task and search suggestion features corresponding to the search suggestion task; and obtains search processing results based on the target city location features and the search suggestion features; and retrieves a list of search suggestions with the highest matching degree to the search processing results from a pre-set search suggestion library and displays the search suggestion list on the search interface. This process replaces the separate processing method of multiple individually modeled systems, using a multi-task search model to simultaneously process search information to obtain search processing results and provide search suggestions to the user. This reduces the training, development, and maintenance costs of multiple models, shortens system response time, and improves the user experience. Attached Figure Description

[0077] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0078] Figure 1a This is a possible illustration of a map search scenario;

[0079] Figure 1b This is an illustration of another possible map search scenario;

[0080] Figure 2 A schematic flowchart illustrating a location-based search processing method provided in an embodiment of this application;

[0081] Figure 3 This is a schematic diagram illustrating the process of training a multi-task search model in an embodiment of this application;

[0082] Figure 4a A schematic diagram illustrating the structure of separate modeling for urban decision-making models and search suggestion models in related technologies;

[0083] Figure 4b This is a schematic diagram of the initial neural network structure in an embodiment of this application;

[0084] Figure 5 A flowchart illustrating another location-based search processing method provided in an embodiment of this application;

[0085] Figure 6 A schematic diagram of the structure of a location-based search processing device provided in an embodiment of this application;

[0086] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application;

[0087] Figure 8 This is a block diagram of a terminal device provided in an exemplary embodiment of this application.

[0088] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0089] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0090] To facilitate understanding of the embodiments of this application, the following terms will be explained before describing the embodiments of this application in detail:

[0091] City-based search decision-making: In the current map search scenario, the system defaults to using city-level administrative divisions as the lowest granularity of the search range. Furthermore, if the user does not actively specify a search city, the system typically uses the user's current city as the search city. For example, a user in Beijing entering "cinema" will return results for cinemas in Beijing. Figure 1a As shown. However, some users may not be targeting their current city and need to search across cities. For example, a user searching for "Oriental Pearl Tower" in Beijing might actually be searching for "Oriental Pearl Radio & TV Tower" in Shanghai. Figure 1b As shown above, the process of deciding which city to switch to is the city-switching decision-making task.

[0092] Search suggestion task: Search suggestions are an important module in search engines. They typically recommend related terms based on prefixes entered by the user, helping them quickly input search keywords and improving search efficiency and accuracy. Search suggestions can be generated based on semantic expansion terms of the query, user search history, high-frequency search terms, and trending events.

[0093] In related technologies, city redirection decision models and search suggestion models are deployed together in the search system. This allows the system to determine whether a city redirection decision is needed based on the user's search input and the user's current location information. This enables the system to obtain the target location of the user's search content and then use the search suggestion model to present content related to the search content to the user from the target location for the user to choose from.

[0094] For city-hopping decision-making tasks, the Learning To Rank (LTR) approach is primarily used to address these problems. This treats the decision-making task as a classification task in machine learning, using cross-entropy as the loss function L = -y. i log(p i )-(1-y i log(1-p) iTrain an LTR ranking model offline, where L represents the loss, y i p is the actual category label for the i-th task. i This is the classification probability of the i-th task predicted by the model. The main process in practical applications is as follows: obtain a set of candidate cities by recalling Points of Interest (POI) results through query terms; construct features of different dimensions (such as query-related features, candidate city-related features, POI set features recalled by candidate cities, matching features between query terms and candidate cities, etc.), and use the LTR model to score and rank the candidate cities. The city with the highest score is the city for the jump decision.

[0095] For search suggestion tasks, common solutions include: Trie tree-based approaches: all search terms are placed in a Trie tree, and as the user inputs each character, words containing that prefix are searched from the Trie tree as search suggestions; alternatively, sequence-to-sequence (seq2seq) deep learning models are used: this model treats the input and output as a sequence. An encoder vectorizes the input words, and a decoder decodes the input vectors to obtain the final output words; another approach is LTR ranking, which constructs features for the query term and candidate words, uses a machine learning model to score the candidate words, and obtains the final search suggestion ranking.

[0096] It is evident that the relevant technologies have the following technical problems: Maintaining offline training processes for two sets of models for different tasks, as well as online logic development modules, can easily lead to high development time and costs; learning a single model task is prone to overfitting, lacking good generalization ability and easily causing data overfitting; furthermore, the two tasks may share similar patterns and rules in certain cases, and separate training will prevent the fusion and sharing of this information, resulting in a lack of information fusion; additionally, if the two models are trained separately, the failure of one task will cause problems in the judgment of subsequent tasks. For example, the failure of the city decision-making task will affect the search suggestions, making them more inclined towards incorrect cities, leading to a domino effect of errors.

[0097] In view of this, embodiments of this application provide a location-based search processing method, apparatus, device, and storage medium. In response to a user's location search request, search information is input into a preset multi-task search model. The search information is processed simultaneously based on a city jump decision task and a search suggestion task within the multi-task search model to obtain target city location features corresponding to the city jump decision task and search suggestion features corresponding to the search suggestion task. Based on the target city location features and the search suggestion features, search processing results are obtained. From a preset search suggestion library, a list of search suggestions with the highest matching degree to the search processing results is obtained and displayed on the search interface. This process replaces the separate processing method of multiple individual models. By using a multi-task search model to simultaneously process search information to obtain search processing results and provide search suggestions to the user, the training, development, and maintenance costs of multiple models can be reduced, system response time can be shortened, and the user experience can be improved.

[0098] The embodiments of this application will be explained below in conjunction with application scenarios. The location-based search processing method provided in the embodiments of this application can be applied to various search engines and applications, such as map applications, life service applications, etc. More specifically, it can be applied to terminal devices with high-precision map apps installed. The following description focuses on the terminal device as the execution subject of the method provided in the embodiments of this application.

[0099] 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. Furthermore, the collection, use and processing of the relevant data must comply with relevant laws, regulations and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0100] The technical solution of this application and how it solves the above-mentioned technical problems will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that these specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0101] Please refer to Figure 2 , Figure 2 The flowchart of a location-based search processing method provided in this application embodiment includes steps S201-S203.

[0102] Step S201: In response to the user's location search request, input the search information into a preset multi-task search model; wherein, the multi-task search model is a model obtained by jointly training multiple tasks, including a city jump decision task and a search suggestion task.

[0103] Understandably, in response to, is used to indicate the conditions or states on which the operation being performed depends. When the conditions or states on which it depends are met, one or more operations being performed can be performed in real time or with a set delay. Unless otherwise specified, there is no restriction on the order in which the multiple operations being performed are executed.

[0104] In one implementation, a user can enter search information in a search interface (such as the search interface of a map search app) to trigger a location search request. The terminal device responds to the user's location search request by inputting the search information into a pre-set multi-task search model.

[0105] The pre-defined multi-task search model in this embodiment is a model jointly trained using multiple tasks, including city-hopping decision-making tasks and search suggestion tasks. In some embodiments, it may also include other related tasks. Compared to city-hopping decision-making task models or search suggestion task models in related technologies, it only requires one model to train multiple tasks simultaneously, reducing system response latency. Furthermore, its training cost is also lower compared to the training process of multiple models.

[0106] In one implementation, the training of a multi-task search model can utilize joint deep learning. This involves jointly training multiple tasks within a single network model, instead of training multiple separate models to handle different tasks. Specifically, this embodiment considers that city analysis and search suggestion tasks involve a wealth of shared input information (such as current city information, historical features of candidate cities, distances, and query term features). Within a single model, shared features across multiple tasks can be learned through a shared underlying neural network, thereby improving the model's generalization ability and efficiency. Furthermore, the model can achieve joint learning of model parameters by jointly training different tasks, thus avoiding overfitting and reducing training time.

[0107] Step S202: The search information is processed simultaneously based on the city jump decision task and the search suggestion task in the multi-task search model to obtain the target city location features corresponding to the city jump decision task and the search suggestion features corresponding to the search suggestion task; and the search processing result is obtained based on the target city location features and the search suggestion features.

[0108] In related technologies, user search information is transmitted sequentially or simultaneously to a city navigation decision-making task model and a search suggestion task model. These two independent models are then used to process the search information accordingly. On the one hand, the training time and cost of independent models are higher, and the different response times of the two models for task processing may affect the overall system response time. On the other hand, if the decision of one task fails, it will cause problems in the judgment of subsequent tasks. For example, the failure of the city decision-making task will cause the search suggestion results to be more inclined to the wrong city, resulting in a domino-like chain reaction of errors, which in turn affects the accuracy of the model.

[0109] In this embodiment, the terminal device only needs to directly input the search information into the multi-task search model. The city jump decision task and search suggestion task in the multi-task search model are used to process the search information simultaneously to obtain the search processing results corresponding to the target city location features and search suggestion features. Since the multi-task search model is trained on multiple tasks simultaneously, it does not need to build two or more network models, resulting in lower training time and cost. During the use of the model, since the tasks are processed simultaneously, there will be no chain reaction of errors, and the processing results are more accurate.

[0110] In one implementation, based on the target city location features and the search suggestion features, the search processing results are obtained. The final search processing result can be obtained by combining the weights corresponding to different features. Specifically, in the multi-task search model, the city jump decision task and the search suggestion task have corresponding feature weights (which may correspond to the first weight and the second weight mentioned later). The search processing result is obtained by summing the product of the target city location features and the first weight with the product of the search suggestion features and the second weight.

[0111] Step S203: Obtain the list of search suggestions with the highest matching degree with the search processing result from the preset search suggestion library, and display the list of search suggestions on the search interface.

[0112] It should be noted that the preset search suggestion library can be predetermined based on existing technologies and practical applications. For example, it can be built using a large amount of historical search data, containing a list of search suggestions corresponding to the results of historical search processing. This list of search suggestions includes one or more suggested search addresses.

[0113] Optionally, the multi-task search model can be trained by incorporating a shared low-level feature extractor, thereby avoiding redundant training and reducing training time and computational resource consumption. Figure 3 As shown, the following steps may be included:

[0114] Step S301: Obtain the training dataset, which includes a city jump decision task sample set and a search suggestion task sample set.

[0115] In this embodiment, only one training dataset is needed to train two tasks simultaneously, eliminating the need to collect multiple training datasets and reducing data collection costs.

[0116] In this embodiment, obtaining the training dataset may include the following steps:

[0117] Obtain historical behavior log data for at least one user, and obtain several data samples based on the historical behavior log data;

[0118] Each data sample is classified according to its task sample type, and the city jump decision task sample set and search suggestion task sample set are obtained based on the classification results to obtain the training dataset.

[0119] It is understood that the data samples in this embodiment include positive samples and negative samples.

[0120] Specifically, two types of (sub)task samples can be collected based on user behavior log data. Session-level sample collection can be used to improve data acquisition efficiency. This means that a set of samples constitutes a single user search behavior, including the user's search terms, click results, city switching results, dwell time, etc. This information is considered as information from a single session and is therefore called a Session-level sample, as shown in Table 1 below:

[0121] Table 1

[0122]

[0123]

[0124] Step S302: Extract sample features based on the training dataset. The sample features include common features, city analysis-related input features, and search suggestion-related input features.

[0125] After obtaining the training dataset, since the sample data in the training dataset may also contain other data, in order to improve the accuracy of model training, this embodiment first reconstructs the features corresponding to the samples in the training dataset to obtain three types of features in the samples: common features, city analysis-related input features, and search suggestion-related input features. Optionally, user behavior logs and feature service logs can be used to reconstruct the features corresponding to the samples.

[0126] Specifically, the training dataset carries identification information for each data sample, and the identification information includes at least one of the following: user identifier, terminal identifier, and log generation time;

[0127] The above process of extracting sample features based on the training dataset may include the following steps:

[0128] For each sample in the training dataset, obtain the service features corresponding to the identification information of the sample from the feature service log data;

[0129] According to the preset feature classification reference table, the service features are classified to obtain sample features including public features, city analysis-related input features, and search suggestion-related input feature types.

[0130] In this embodiment, the feature service log data refers to the feature data recorded in the feature server for each sample. Specifically, the service features are written directly to the feature server, and the information of the feature server can be persisted to disk or stored in a Hadoop cluster for convenient subsequent data preprocessing.

[0131] Understandably, user behavior logs are used to construct sample labels, but a model trained solely on labels is insufficient for model training; additional features corresponding to the samples are required. The system stores the feature values ​​generated by each user request on a feature server to form feature service log data. Finally, by associating the training dataset obtained from the behavior logs with the information from the feature server (association between identification information), the final sample (label + feature) is obtained. Furthermore, those skilled in the art can pre-set a feature classification reference table based on practical applications and existing technologies; for example, the Oriental Pearl Tower can be classified as a city-related input feature for analysis.

[0132] In one implementation, a feature classification reference table is used to categorize features into public features, city analysis-related input features, and search suggestion-related input features. Examples of features are shown in Table 2 below:

[0133] Table 2

[0134]

[0135] Step S303: Train an initial neural network based on the sample features to obtain the multi-task search model; wherein, the initial neural network includes: an initial city jump decision task and an initial search suggestion task.

[0136] In this embodiment, information fusion can be achieved by leveraging the shared patterns and rules between the two tasks in some search cases. Models trained separately cannot fuse and share this information. Because the training dataset in this application is generic, the model can fully utilize the shared patterns and rules in both tasks by using common features, thereby improving the model's performance.

[0137] In one implementation, compared to, Figure 4a In related technologies, two separate network models are used to train two tasks. For example... Figure 4b As shown, the initial neural network in this embodiment includes: an input layer 41; a data sharing layer 42 connected to the input layer 41; a city analysis feature extraction layer 43 and a search suggestion feature extraction layer 44 connected to the data sharing layer 42 respectively; and an output layer 45 connected to the city analysis feature extraction layer 43 and the search suggestion feature extraction layer 44 respectively.

[0138] In this embodiment, the data sharing layer contains several hidden layers. These hidden layers are used to train the initial city jump decision task and the initial search suggestion task simultaneously using common features. The connections between the hidden layers can be represented as weights.

[0139] Step S303 above, which trains an initial neural network based on the sample features to obtain the multi-task search model, may include the following steps:

[0140] The sample features are input into the input layer;

[0141] If the sample feature is a common feature, the input layer is triggered to transmit the common feature to the data sharing layer; and / or, if the sample feature is a city analysis-related input feature, the input layer is triggered to transmit the city analysis-related input feature to the city analysis feature extraction layer; and / or, if the sample feature is a search suggestion-related input feature, the input layer is triggered to transmit the search suggestion-related input feature to the search suggestion feature extraction layer.

[0142] The data sharing layer is triggered to simultaneously train the initial city jump decision task and the initial search suggestion task based on the common features, and the training results are sent to the city analysis feature extraction layer and the search suggestion feature extraction layer, respectively.

[0143] The city analysis feature extraction layer is triggered to obtain the target city location features based on the training processing results and the city analysis-related input features; and the search suggestion feature extraction layer is triggered to obtain the search suggestion features based on the training processing results and the search suggestion-related input features.

[0144] The output layer is triggered to output the trained target city location features and the trained search suggestion features; and the loss value between the trained target city location features and the trained search suggestion features and the preset target features is obtained;

[0145] The first weight corresponding to the initial city jump decision task and the second weight corresponding to the initial search suggestion task are adjusted according to the loss value to obtain the trained city jump decision task and the trained search suggestion task.

[0146] Based on the trained city jump decision task and the trained search suggestion task, a trained deep neural network is obtained.

[0147] In this embodiment, the initial neural network model can be a deep neural network (DNN) model. In other embodiments, other neural network models can also be used. This embodiment does not impose any other limitations.

[0148] It should be noted that the preset target feature in this embodiment can be the real feature corresponding to the training data, such as the real feature corresponding to the output result of each group of samples; or it can be the predicted feature in the prediction dataset, which is a standard value. The loss value obtained between the trained target city location feature and the trained search suggestion feature and the preset target feature can be calculated by summing the products of the trained city analysis-related input feature and the trained search suggestion-related input feature with the first weight value and the second weight, respectively, and then using this sum with the real feature. The calculation of the loss is a prior art technique and will not be elaborated further here.

[0149] It should be noted that in this embodiment, the first weight and the second weight are parameters of the neural network (the neural network parameters can be understood to include weight parameters, and by training the weight parameters, the prediction accuracy of the neural network can be improved). The first weight and the second weight are used to distinguish similar objects and have no other specific meaning. The first weight and the second weight can have the same weight value or different weight values.

[0150] As can be seen, compared with the models provided by related technologies, the multi-task model adopted in this embodiment shares the underlying feature weights and can simultaneously train the first and second weights corresponding to multiple tasks. This prevents the model from overfitting due to excessive learning of a single task. Furthermore, because the underlying feature weights are shared, the model can fully utilize the same patterns and rules in both tasks, improving the model's performance.

[0151] In one implementation, training an initial neural network by determining training parameters to improve the performance and effectiveness of model learning may further include the following steps:

[0152] Determine the training parameters in the initial neural network, wherein the training parameters include the learning rate and / or the processing batch;

[0153] The step of adjusting the first weight corresponding to the initial city jump decision task and adjusting the second weight corresponding to the initial search suggestion task based on the loss value includes:

[0154] Based on the training parameters, the first weight corresponding to the initial city jump decision task and the second weight corresponding to the initial search suggestion task are adjusted according to the loss value.

[0155] Furthermore, after completing the training and parameter tuning (weight adjustment) of the model, the model can be evaluated offline (using existing model evaluation methods). If the evaluation result is poor, the training parameters (i.e., hyperparameters) of the initial neural network are redefined, and the initial neural network is retrained using these training parameters to improve the accuracy of the model.

[0156] Furthermore, after completing model training and parameter tuning, A / B experiments can be conducted to improve model accuracy. A / B experiments, also known as AB testing, AB bucketing, etc., involve creating two (A / B) or more (A / B / n) versions of the model. Simultaneously, users with similar or identical user compositions randomly access these versions, collecting behavioral data from each group. Finally, the best version is analyzed and adopted. After a period of experimentation, the results of the online A / B experiment are analyzed to see if the experimental group shows significant positive benefits. If not, the reasons for failure are analyzed, and the model is optimized from the perspective of samples or features, and the experiment is repeated. If positive benefits are observed, the experimental portion of the traffic is fully utilized, and the experiment ends.

[0157] The multi-task search model in this embodiment uses the same model to train two strongly related tasks simultaneously, and shares feature weights at the underlying level, which reduces the overfitting problem of a single task and improves the generalization ability of the model. Furthermore, replacing the previous two models with one model reduces the cost of sample generation, feature construction and model training by half, allowing the model to be iteratively optimized more quickly.

[0158] Please refer to Figure 5 , Figure 5This is a flowchart illustrating another location-based search processing method provided in this application embodiment. Based on the above embodiment, considering that the model accumulates more search data after a period of use, in order to further improve the accuracy of model prediction, this embodiment performs periodic updates on the multi-task model and uses the updated model for location search processing. In addition to the above steps S201-S203, the method of this embodiment may also include the following steps:

[0159] Step S501: Update the preset multi-task search model periodically according to a preset cycle.

[0160] It should be noted that those skilled in the art can adapt the preset period by combining existing technology and practical applications, for example, by determining the preset period based on the user's location search frequency.

[0161] Specifically, the periodic update process may include the following steps:

[0162] The log data corresponding to the search processing performed by the multi-task model is written to the search log database in real time, so as to obtain a new training dataset for update processing based on the log database.

[0163] Obtain the duration of the search processing performed by the multi-task model;

[0164] When the duration reaches the calibrated value, the multi-task model is trained using the new training dataset, and the trained multi-task model is used as the updated multi-task training model.

[0165] In this embodiment, the calibration value can be a predetermined value, which can be adaptively set according to user needs. For example, the duration of the multi-task model's search processing is two months, and the model is periodically updated when the two-month period (calibration value) is reached.

[0166] To facilitate understanding of the embodiments of this application, an exemplary embodiment is provided, comprising two parts.

[0167] Part 1: Offline model training, including the following processes S1-S7.

[0168] S1. Data Collection: Two sub-task samples are collected based on user behavior logs. Here, we adopted a session-level sample collection method. That is, a set of samples constituted by a user's search behavior, which includes information such as the user's search terms, click results, city switching results, dwell time, etc. This information is regarded as the information of a session, and is therefore called session-level sample, as shown in Table 1.

[0169] S2. Feature Restoration (or Extraction): Using user behavior logs and feature service logs, restore the features corresponding to the sample;

[0170] S3. Feature Classification: Features are classified into common features, other input features of city distribution, and other input features of search suggestions. Examples of features are shown in Table 2.

[0171] S4, Model Training and Parameter Tuning;

[0172] S5, offline evaluation;

[0173] S6. If the evaluation results are poor, adjust the hyperparameters used for model training and parameter tuning, and retrain the model.

[0174] S7. Save the model:

[0175] Part Two: Online Model Deployment, including the following processes S8-S15.

[0176] S8. Convert the deep learning model to ONNX format;

[0177] S9. Online implementation of Java calling ONNX project code;

[0178] S10. When a user query request arrives, the recall results are obtained and samples are constructed.

[0179] S11. Call the model to predict the candidate results and obtain the corresponding search suggestions and city selection results;

[0180] S12. Return the list of POI results desired by the user;

[0181] S13. After a period of time, analyze the results of the online A / B experiment to see if the experimental group has significant positive benefits.

[0182] S14. If not, analyze the reasons for failure, optimize the model from the perspective of samples or features, and repeat the experiment.

[0183] S15. If so, then the full amount of experimental data will be used, and the experiment will end.

[0184] Furthermore, since the above logs are collected offline, the shortest update cycle for the model is T+1. We can try adding the Flink real-time stream processing framework to the user log collection and analysis module to achieve real-time analysis of user behavior logs, realize hourly updates of the model, and enable rapid iterative updates.

[0185] Furthermore, the historical behavior records of a user group or even an individual can be analyzed to generate corresponding results, thereby obtaining personalized city redirection and search suggestions.

[0186] Compared to related technologies, this embodiment utilizes a multi-task search model under a joint learning framework to achieve location search processing, improving the effectiveness of city jump decisions and search suggestion ranking in map search scenarios, thereby optimizing the city analysis and search suggestion processing process. Furthermore, the multi-task search model shares underlying features, thus avoiding redundant training, reducing training time and computational resource consumption, and improving the efficiency of update iteration.

[0187] Please refer to Figure 6 , Figure 6 A schematic diagram of a location-based search processing device provided in this application embodiment includes an input module 61, a model processing module 62, and a matching module 63, wherein...

[0188] The input module 61 is configured to input search information into a preset multi-task search model in response to a user's location search request; wherein the multi-task search model is a model obtained by jointly training multiple tasks, including a city jump decision task and a search suggestion task.

[0189] The model processing module 62 is configured to simultaneously process the search information based on the city jump decision task and the search suggestion task in the multi-task search model, so as to obtain the target city location features corresponding to the city jump decision task and the search suggestion features corresponding to the search suggestion task; and obtain the search processing results based on the target city location features and the search suggestion features.

[0190] The matching module 63 is configured to retrieve a list of search suggestions that have the highest matching degree with the search processing result from a preset search suggestion library, and display the list of search suggestions on the search interface.

[0191] In one implementation, the module for training the multi-task search model includes:

[0192] The dataset acquisition module is configured to acquire a training dataset, which includes a city jump decision task sample set and a search suggestion task sample set.

[0193] The feature extraction module is configured to extract sample features based on the training dataset, the sample features including common features, city analysis-related input features, and search suggestion-related input features;

[0194] The training module is configured to train an initial neural network based on the sample features to obtain the multi-task search model; wherein the initial neural network includes an initial city jump decision task and an initial search suggestion task.

[0195] In one implementation, the dataset acquisition module includes:

[0196] The log acquisition unit is configured to acquire historical behavior log data corresponding to at least one user, and acquire several data samples based on the historical behavior log data.

[0197] The classification unit is configured to classify each data sample according to the task sample type, and obtain the city jump decision task sample set and the search suggestion task sample set based on the classification results to obtain the training dataset.

[0198] In one implementation, the training dataset carries identification information for each data sample, the identification information including at least one of the following: user identifier, terminal identifier, and log generation time;

[0199] The feature extraction module includes:

[0200] The service feature acquisition unit is configured to acquire service features corresponding to the identification information of each sample from the feature service log data for each sample in the training dataset.

[0201] The feature classification unit is configured to classify the service features according to a preset feature classification reference table to obtain sample features including public features, city analysis-related input features, and search suggestion-related input feature types.

[0202] In one embodiment, the initial neural network includes: an input layer; a data sharing layer connected to the input layer; a city analysis feature extraction layer and a search suggestion feature extraction layer connected to the data sharing layer, respectively; and an output layer connected to the city analysis feature extraction layer and the search suggestion feature extraction layer, respectively.

[0203] The training module includes:

[0204] An input unit is configured to input the sample features into the input layer;

[0205] The first triggering unit is configured to, if the sample feature is a common feature, trigger the input layer to transmit the common feature to the data sharing layer; and / or, if the sample feature is a city analysis-related input feature, trigger the input layer to transmit the city analysis-related input feature to the city analysis feature extraction layer; and / or, if the sample feature is a search suggestion-related input feature, trigger the input layer to transmit the search suggestion-related input feature to the search suggestion feature extraction layer.

[0206] The second triggering unit is configured to trigger the data sharing layer to simultaneously train the initial city jump decision task and the initial search suggestion task based on the public features, and send the training results to the city analysis feature extraction layer and the search suggestion feature extraction layer respectively.

[0207] The third triggering unit is configured to trigger the city analysis feature extraction layer to obtain the target city location features based on the training processing results and the city analysis-related input features; and to trigger the search suggestion feature extraction layer to obtain the search suggestion features based on the training processing results and the search suggestion-related input features.

[0208] The fourth triggering unit is configured to trigger the output layer to output the trained target city location features and the trained search suggestion features; and to obtain the loss value between the trained target city location features and the trained search suggestion features and the preset target features;

[0209] The weight adjustment unit is configured to adjust the first weight corresponding to the initial city jump decision task and the second weight corresponding to the initial search suggestion task according to the loss value, so as to obtain the trained city jump decision task and the trained search suggestion task.

[0210] The acquisition unit is configured to obtain a trained deep neural network based on the trained city jump decision task and the trained search suggestion task.

[0211] In one implementation, it further includes:

[0212] A parameter determination module is configured to determine training parameters in the initial neural network, wherein the training parameters include a learning rate and / or a batch size.

[0213] The weight adjustment unit is specifically configured to adjust the first weight corresponding to the initial city jump decision task and the second weight corresponding to the initial search suggestion task based on the training parameters and the loss value.

[0214] In one embodiment, the device further includes:

[0215] The update module is configured to periodically update the preset multi-task search model according to a preset cycle.

[0216] The periodic update process includes:

[0217] The log data corresponding to the search processing performed by the multi-task model is written to the search log database in real time, so as to obtain a new training dataset for update processing based on the log database.

[0218] Obtain the duration of the search processing performed by the multi-task model;

[0219] When the duration reaches the calibrated value, the multi-task model is trained using the new training dataset, and the trained multi-task model is used as the updated multi-task training model.

[0220] For relevant instructions, please refer to the corresponding text. Figures 2-5 The relevant descriptions and effects of the steps in the corresponding embodiments are understood, and will not be elaborated on here.

[0221] This application also provides an electronic device, such as... Figure 7 As shown, it includes: a memory 71 and a processor 72;

[0222] The memory 71 stores computer-executed instructions;

[0223] The processor 72 executes computer execution instructions stored in the memory 71, causing the electronic device to perform the location-based search processing method, wherein the memory 71 and the processor 72 are connected via a bus 73.

[0224] For relevant instructions, please refer to the corresponding text. Figures 2-5 The relevant descriptions and effects of the steps in the corresponding embodiments are understood, and will not be elaborated on here.

[0225] According to another aspect of this application, a computer-readable storage medium is provided, wherein computer-executable instructions are stored therein, which, when executed by a processor, are used to implement the location-based search processing method described above.

[0226] The computer-readable storage medium can be ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0227] For relevant instructions, please refer to the corresponding text. Figures 2-5 The relevant descriptions and effects of the steps in the corresponding embodiments are understood, and will not be elaborated on here.

[0228] One embodiment of this application provides a computer program product, including a computer program that, when executed by a processor, implements this application. Figures 2-5 The location-based search processing method provided in any of the corresponding embodiments.

[0229] Figure 8This is a block diagram illustrating an exemplary embodiment of the present application of a terminal device 800, which may be a mobile phone, computer, digital broadcasting terminal, messaging device, game console, tablet device, medical device, fitness device, personal digital assistant, etc.

[0230] The terminal device 800 may include one or more of the following components: processing component 802, memory 804, power supply component 806, multimedia component 808, audio component 810, input / output (I / O) interface 812, sensor component 814, and communication component 816.

[0231] Processing component 802 typically controls the overall operation of terminal device 800, such as operations associated with display, telephone calls, data communication, camera operation, and recording. Processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the methods described above. Furthermore, processing component 802 may include one or more modules to facilitate interaction between processing component 802 and other components. For example, processing component 802 may include a multimedia module to facilitate interaction between multimedia component 808 and processing component 802.

[0232] Memory 804 is configured to store various types of data to support operation on terminal device 800. Examples of this data include instructions for any application or method operating on terminal device 800, contact data, phonebook data, messages, pictures, videos, etc. Memory 804 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0233] Power supply component 806 provides power to various components of terminal device 800. Power supply component 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to terminal device 800.

[0234] Multimedia component 808 includes a screen that provides an output interface between terminal device 800 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touchscreen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors may sense not only the boundaries of touch or swipe actions but also the duration and pressure associated with the touch or swipe operation. In some embodiments, multimedia component 808 includes a front-facing camera and / or a rear-facing camera. When terminal device 800 is in an operating mode, such as a shooting mode or video mode, the front-facing camera and / or rear-facing camera may receive external multimedia data. Each front-facing camera and rear-facing camera may be a fixed optical lens system or have focal length and optical zoom capabilities.

[0235] Audio component 810 is configured to output and / or input audio signals. For example, audio component 810 includes a microphone (MIC) configured to receive external audio signals when terminal device 800 is in an operating mode, such as call mode, recording mode, and voice recognition mode. The received audio signals may be further stored in memory 804 or transmitted via communication component 816. In some embodiments, audio component 810 also includes a speaker for outputting audio signals.

[0236] I / O interface 812 provides an interface between processing component 802 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to, home buttons, volume buttons, power buttons, and lock buttons.

[0237] Sensor assembly 814 includes one or more sensors for providing status assessments of various aspects of terminal device 800. For example, sensor assembly 814 can detect the on / off state of terminal device 800, the relative positioning of components such as the display and keypad of terminal device 800, changes in the position of terminal device 800 or a component of terminal device 800, the presence or absence of user contact with terminal device 800, the orientation or acceleration / deceleration of terminal device 800, and temperature changes of terminal device 800. Sensor assembly 814 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. Sensor assembly 814 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, sensor assembly 814 may also include an accelerometer, a gyroscope, a magnetometer, a pressure sensor, or a temperature sensor.

[0238] Communication component 816 is configured to facilitate wired or wireless communication between terminal device 800 and other devices. Terminal device 800 can access wireless networks based on communication standards, such as WiFi, 3G, 4G, 5G, or other standard communication networks, or combinations thereof. In one exemplary embodiment, communication component 816 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, communication component 816 also includes a near-field communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on radio frequency identification (RFID) technology, Infrared Data Association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0239] In an exemplary embodiment, the terminal device 800 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the functions described in this application. Figures 2-5 The method provided in any of the corresponding embodiments.

[0240] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 804 including instructions, which can be executed by a processor 820 of a terminal device 800 to perform the above-described method. For example, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0241] This application also provides a non-transitory computer-readable storage medium, which, when the instructions in the storage medium are executed by the processor of a terminal device, enables the terminal device 800 to perform the above-described embodiments of this application. Figures 2-5 The method provided in any of the corresponding embodiments.

[0242] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0243] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the application disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0244] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A location-based search processing method, characterized in that, include: In response to a user's location search request, the search information is input into a pre-defined multi-task search model; The search information is processed simultaneously based on the city jump decision task and the search suggestion task in the multi-task search model to obtain the target city location features corresponding to the city jump decision task and the search suggestion features corresponding to the search suggestion task. Based on the target city location features and the search suggestion features, the search processing results are obtained; In the preset search suggestion library, obtain the list of search suggestions that match the search results most closely, and display the list of search suggestions on the search interface; The methods for training the multi-task search model include: Obtain a training dataset, which includes a sample set of city jump decision tasks and a sample set of search suggestion tasks; Sample features are extracted based on the training dataset. These sample features include common features, city analysis-related input features, and search suggestion-related input features. The initial neural network is trained based on the sample features to obtain the multi-task search model; wherein, the multi-task search model is a model obtained by jointly training the city jump decision task and the search suggestion task using the common features using the initial neural network containing a data sharing layer.

2. The method according to claim 1, characterized in that, The acquisition of the training dataset includes: Obtain historical behavior log data for at least one user, and obtain several data samples based on the historical behavior log data; Each data sample is classified according to its task sample type, and the city jump decision task sample set and search suggestion task sample set are obtained based on the classification results to obtain the training dataset.

3. The method according to claim 2, characterized in that, The training dataset carries identification information for each data sample, and the identification information includes at least one of the following: user identifier, terminal identifier, and log generation time; The extraction of sample features based on the training dataset includes: For each sample in the training dataset, obtain the service features corresponding to the identification information of the sample from the feature service log data; According to the preset feature classification reference table, the service features are classified to obtain sample features including public features, city analysis-related input features, and search suggestion-related input feature types.

4. The method according to any one of claims 1-3, characterized in that, The initial neural network includes: an input layer; a data sharing layer connected to the input layer; a city analysis feature extraction layer and a search suggestion feature extraction layer connected to the data sharing layer, respectively; and an output layer connected to the city analysis feature extraction layer and the search suggestion feature extraction layer, respectively. The process of training an initial neural network based on the sample features to obtain the multi-task search model includes: The sample features are input into the input layer; If the sample feature is a common feature, the input layer is triggered to transmit the common feature to the data sharing layer; and / or, if the sample feature is a city analysis-related input feature, the input layer is triggered to transmit the city analysis-related input feature to the city analysis feature extraction layer; and / or, if the sample feature is a search suggestion-related input feature, the input layer is triggered to transmit the search suggestion-related input feature to the search suggestion feature extraction layer. The data sharing layer is triggered to simultaneously train the initial city jump decision task and the initial search suggestion task based on the common features, and the training results are sent to the city analysis feature extraction layer and the search suggestion feature extraction layer respectively. The city analysis feature extraction layer is triggered to obtain the target city location features based on the training processing results and the city analysis-related input features; and the search suggestion feature extraction layer is triggered to obtain the search suggestion features based on the training processing results and the search suggestion-related input features. The output layer is triggered to output the trained target city location features and the trained search suggestion features; and the loss value between the trained target city location features and the trained search suggestion features and the preset target features is obtained; The first weight corresponding to the initial city jump decision task and the second weight corresponding to the initial search suggestion task are adjusted according to the loss value to obtain the trained city jump decision task and the trained search suggestion task. Based on the trained city jump decision task and the trained search suggestion task, a trained deep neural network is obtained.

5. The method according to claim 4, characterized in that, Also includes: Determine the training parameters in the initial neural network, wherein the training parameters include the learning rate and / or the processing batch; The step of adjusting the first weight corresponding to the initial city jump decision task and adjusting the second weight corresponding to the initial search suggestion task based on the loss value includes: Based on the training parameters, the first weight corresponding to the initial city jump decision task and the second weight corresponding to the initial search suggestion task are adjusted according to the loss value.

6. The method according to claim 1, characterized in that, Also includes: The preset multi-task search model is periodically updated according to a preset cycle. The periodic update process includes: The log data corresponding to the search processing performed by the multi-task search model is written to the search log database in real time, so as to obtain a new training dataset for update processing based on the log database. Obtain the duration of the search processing performed by the multi-task search model; When the duration reaches the calibrated value, the multi-task search model is trained using the new training dataset, and the trained multi-task search model is used as the updated multi-task search model.

7. A location-based search processing device, characterized in that, include: The input module is configured to respond to a user's location search request by inputting search information into a pre-defined multi-task search model; The model processing module is configured to process the search information simultaneously based on the city jump decision task and the search suggestion task in the multi-task search model, so as to obtain the target city location features corresponding to the city jump decision task and the search suggestion features corresponding to the search suggestion task. Based on the target city location features and the search suggestion features, the search processing results are obtained; The matching module is configured to retrieve a list of search suggestions that have the highest matching degree with the search processing result from a preset search suggestion library, and display the list of search suggestions on the search interface; The module for training the multi-task search model includes: The dataset acquisition module is configured to acquire a training dataset, which includes a city jump decision task sample set and a search suggestion task sample set. The feature extraction module is configured to extract sample features based on the training dataset, the sample features including common features, city analysis-related input features, and search suggestion-related input features; The training module is configured to train an initial neural network based on the sample features to obtain the multi-task search model; wherein the multi-task search model is a model obtained by jointly training an initial neural network containing a data sharing layer and using the common features for the city jump decision task and the search suggestion task.

8. An electronic device, characterized in that, include: Memory and processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the electronic device to perform the location-based search processing method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the location-based search processing method as described in any one of claims 1 to 6.

Images