City search intention recognition method, device and equipment and storage medium

Abstract

The application belongs to the technical field of deep learning, and provides a city retrieval intention recognition method, device, equipment and storage medium, wherein the method comprises the following steps: acquiring address retrieval information of a user; matching the address retrieval information with a city dictionary; when the matched cities are more than one, matching the address retrieval information with a special word dictionary; when the matched cities are more than one, matching the address retrieval information with a filtered city dictionary to obtain a standard city; predicting the address retrieval information based on a multi-label FastText model to generate a sequence list; and screening a target city based on the standard city and the sequence list, and matching an address belonging to the target city to recommend to the user, so as to avoid recalling too many redundant results, reduce the pressure of sorting, and improve the accuracy of retrieval results.

Description

Technical Field

[0001] This application relates to the field of deep learning technology, and more specifically, to a method, apparatus, device, and storage medium for identifying urban search intent. Background Technology

[0002] With the development of smart devices and mobile internet technology, new modes of transportation that combine with the internet have also flourished. To better meet user needs, accurately identifying the user's desired destination is crucial for understanding search intent.

[0003] Current search intent recognition solutions mainly perform precise matching of user-input location terms across the country, then score the results based on existing results in the database, and return the result with the highest similarity score. However, precise matching of destinations across the country can easily recall too many redundant results, putting too much pressure on the ranking and resulting in inaccurate search results. Summary of the Invention

[0004] The main objective of this application is to provide a method, apparatus, device, and storage medium for identifying urban search intent, so as to avoid recalling too many redundant results, reduce the pressure of sorting, and improve the accuracy of search results.

[0005] To achieve the above-mentioned objectives, this application provides a method for identifying urban search intent, comprising:

[0006] Obtain the user's address search information;

[0007] The address retrieval information is matched with the city dictionary;

[0008] When more than one city is matched, the address retrieval information is matched with a special word dictionary; wherein, the special word dictionary includes cities and corresponding special words representing the cities;

[0009] When more than one city is matched, the address retrieval information is matched using a filtered city dictionary to obtain standard cities; wherein, the filtered city dictionary is a dictionary with the administrative division suffix removed.

[0010] The address retrieval information is predicted based on the multi-label FastText model to generate a sequence list; wherein, the sequence list includes multiple candidate cities and the predicted probability of each candidate city;

[0011] Based on the standard cities and the sequence list, target cities are filtered out, and addresses belonging to the target cities are matched and recommended to the user.

[0012] Furthermore, before matching the address retrieval information with a special word dictionary, the process also includes:

[0013] Obtain user click data and order data within a preset time period on the freight platform;

[0014] Historical address retrieval information with unique matching cities is summarized from the click data and order data;

[0015] The matching cities and corresponding special words representing the cities are filtered out from the historical address retrieval information to generate a special word dictionary.

[0016] Preferably, the target city is selected based on the standard city and the sequence list, including:

[0017] Iterate through the first N candidate cities in the sequence list; where N is a positive integer;

[0018] Determine whether the candidate city encountered during the iteration is ranked first in the sequence list;

[0019] If so, add the candidate city to the city list;

[0020] If not, determine whether the distance between the candidate cities traversed and the user's city is less than the preset distance and whether the corresponding predicted probability is greater than or equal to the first preset threshold.

[0021] If so, add the candidate city to the first candidate list;

[0022] If not, determine whether the candidate cities encountered are the user's city;

[0023] If so, add the candidate city to the city list;

[0024] If not, determine whether the candidate cities encountered are standard cities;

[0025] If so, add the candidate city to the second candidate list;

[0026] The target city is selected based on the first candidate list, the second candidate list, and the city list.

[0027] Preferably, the target city is selected based on the first candidate list, the second candidate list, and the city list, including:

[0028] Filter out the candidate cities that are closest to the user's city from the first candidate list and add them to the city list;

[0029] Add all candidate cities from the second candidate list to the city list;

[0030] Determine whether the number of candidate cities in the city list is greater than or equal to 3;

[0031] If so, select the target city from the city list.

[0032] Furthermore, after determining whether the number of candidate cities in the city list is greater than or equal to 3, the process also includes:

[0033] If not, iterate through the candidate cities in the sequence list from the Nth to the Mth position; where M is a positive integer greater than N.

[0034] Determine whether the candidate cities traversed are the closest to the user's city and have a predicted probability greater than a second preset threshold.

[0035] If so, add the candidate city to the city list.

[0036] Furthermore, before generating the sequence list by predicting the address retrieval information based on the multi-label FastText model, the following steps are also included:

[0037] Acquire training data; wherein the training data is fused data consisting of local data and cross-city data, and the fused data includes user click data, order data, PO I data and address data;

[0038] The FastText model is trained using the training data in a multi-label manner to obtain a multi-label FastText model.

[0039] Furthermore, after matching the address retrieval information with the city dictionary, the process also includes:

[0040] When a unique city is matched, that city is selected as the target city and recommended to the user.

[0041] This application also provides a city search intent recognition device, the device comprising:

[0042] The acquisition module is used to obtain the user's address retrieval information;

[0043] The first matching module is used to match the address retrieval information with a city dictionary;

[0044] The second matching module is used to match the address retrieval information with a special word dictionary when the number of matched cities is greater than 1; wherein, the special word dictionary includes cities and corresponding special words representing the cities;

[0045] The third matching module is used to match the address retrieval information using a filtered city dictionary when the number of matched cities is greater than 1, in order to obtain standard cities; wherein, the filtered city dictionary is a dictionary with the administrative division suffix removed.

[0046] The prediction module is used to predict the address retrieval information based on the multi-label FastText model and generate a sequence list; wherein, the sequence list includes multiple candidate cities and the predicted probability of each candidate city;

[0047] The filtering module is used to filter target cities based on standard cities and the sequence list, and to recommend addresses belonging to the target cities to users.

[0048] This application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods described above.

[0049] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the methods described above.

[0050] The present application provides a method, apparatus, device, and storage medium for identifying city search intent. It utilizes a city dictionary, a special word dictionary, and a filtered city dictionary to initially determine the city intent of a user's address search information. Simultaneously, it combines a multi-label FastText model to predict the address search information and obtain a sequence list. Based on standard cities and the sequence list, it determines the target city to which the address search information belongs, thus retrieving the destination within the target city. This avoids recalling too many redundant results, reduces the pressure of ranking, and the FastText model trained using a multi-label approach improves the prediction accuracy for each city, narrowing the gap between model training and prediction, thereby improving the accuracy of the search results. Attached Figure Description

[0051] Figure 1 This is a flowchart illustrating a city search intent recognition method according to an embodiment of this application;

[0052] Figure 2 This is a schematic block diagram of a city search intent recognition device according to an embodiment of this application;

[0053] Figure 3 This is a schematic block diagram of the structure of a computer device according to an embodiment of this application.

[0054] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0055] 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.

[0056] This application proposes a method for identifying urban search intent, in which a computer device is executed. The computer device can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or PCs. The computing device can also be a mobile or stationary server. Components of the computing device include, but are not limited to, memory and a processor. The processor and memory are connected via a bus, and a database is used to store data. The computing device also includes an access device that enables the computing device to communicate via one or more networks. Examples of these networks include the Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or combinations of communication networks such as the Internet. Access devices may include one or more of any type of wired or wireless network interface (e.g., a network interface card (NIC)), such as an IEEE 802.11 wireless local area network (WLAN) interface, a Wi-MAX interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a Near Field Communication (NFC) interface, and so on.

[0057] refer to Figure 1 In one embodiment, this application provides a method for identifying urban search intent, the method comprising:

[0058] S11. Obtain the user's address retrieval information;

[0059] S12. Match the address retrieval information with the city dictionary;

[0060] S13. When the number of matched cities is greater than 1, the address retrieval information is matched with a special word dictionary; wherein, the special word dictionary includes cities and corresponding special words representing the cities;

[0061] S14. When the number of matched cities is greater than 1, the address retrieval information is matched using the filtered city dictionary to obtain standard cities; wherein, the filtered city dictionary is a dictionary with the administrative division suffix removed.

[0062] S15. Based on the multi-label FastText model, predict the address retrieval information and generate a sequence list; wherein, the sequence list includes multiple candidate cities and the predicted probability of each candidate city;

[0063] S16. Target cities are filtered based on standard cities and the sequence list, and addresses belonging to the target cities are matched and recommended to the user.

[0064] As described in step S11 above, users can enter address search information in the search bar by manual input or voice input. The address search information may include keywords of the destination, such as street name, city / region, house number, floor, postal code and building name, etc.

[0065] As described in step S12 above, a city dictionary refers to a database or data structure that contains the names and corresponding information of various cities. This dictionary is typically used to store and query information such as city names, countries / regions, postal codes, latitude and longitude, and population sizes worldwide. A city dictionary provides a fast and convenient way to access city information and can be used in areas such as address retrieval, logistics and delivery, travel planning, and weather forecasting. It helps users quickly find information about specific cities without having to traverse the entire database or perform complex searches.

[0066] This embodiment can extract the city name from the address search information and use the city dictionary lookup function to match the extracted city name in the dictionary. If the dictionary contains only the corresponding city name, the city name is returned. For example, if the user's address search information is "No. 188, Huaihai Middle Road, Shanghai", and the city dictionary matching strategy finds only one city, "Shanghai", then the result "Shanghai" is returned directly.

[0067] In one embodiment, after finding the corresponding city name in the city dictionary, other relevant information about the city, such as its country / region, postal code, latitude and longitude, can also be obtained. Based on the matching results, further operations can be performed, such as displaying the query results or locating the address.

[0068] As described in step S13 above, when the matching result includes more than one city, the address retrieval information is matched with a special word dictionary. If the special word dictionary matching strategy can match a unique city, that city is directly returned. This special word dictionary includes cities and corresponding special words representing those cities, such as "Canton Tower" and "Yangcheng" for Guangzhou, and "Hometown of Martial Arts" and "Huang Feihong" for Foshan, etc.

[0069] In one embodiment, before matching the address retrieval information with a special word dictionary, the method further includes:

[0070] Obtain user click data and order data within a preset time period on the freight platform;

[0071] Historical address retrieval information with unique matching cities is summarized from the click data and order data;

[0072] The matching cities and corresponding special words representing the cities are filtered out from the historical address retrieval information to generate a special word dictionary.

[0073] This embodiment can statistically analyze user click and order data on the freight platform within six months, and aggregate historical address search information that appears only in one city. For example, if the historical address search information contains "Yangcheng," since "Yangcheng" only refers to Guangzhou, this historical address search information will be aggregated; if the historical address search information contains "Zhongshan Avenue," since "Zhongshan Avenue" may refer to streets in cities such as Guangzhou and Nanjing, this historical address search information will not be aggregated, to ensure the uniqueness of the aggregated historical address search information and the special word dictionary and avoid errors.

[0074] From each historical address retrieval information, matching cities and corresponding special words representing those cities are filtered out, constructing a special word dictionary (key is the special word, value is the corresponding city). During matching, if the user's address retrieval information is a key in the special word dictionary, the corresponding city is directly returned.

[0075] As described in step S14 above, when more than one city is matched after matching the address retrieval information with the special word dictionary, the filtered city dictionary is used to match the address retrieval information to obtain the standard city. This filtered city dictionary is a dictionary with the administrative division suffix removed. For example, "city / county / district" etc., after removing "Shanghai City", becomes "Shanghai", thus constructing the filtered city dictionary.

[0076] As described in steps S15-S16 above, this embodiment can use the multi-label FastText model in the freight scenario to predict the user's address retrieval information. The FastText model performs inference through forward propagation and returns a sequence list. Each element in the sequence list is a tuple (city and its corresponding prediction probability). That is, the sequence list gives all candidate cities, the prediction probability of each candidate city, and the order of all candidate cities in the sequence list. Then, the target city is selected based on the standard city and the sequence list. For example, the standard city in the sequence list is selected as the target city. Finally, the address of the target city is further matched based on the address retrieval information and recommended to the user so as to retrieve the destination in the target city, avoid recalling too many redundant results, and reduce the pressure of sorting.

[0077] FastText is a word vector-based text classification and encoding model. Developed by Facebook AI Research, it aims to provide efficient text representation and fast text classification. The core idea of ​​FastText is to break down the input text into character-level n-grams and represent each character-level n-gram as a vector. By averaging or concatenating these character-level n-gram vectors, the representation of the entire text can be obtained. This representation is then used for tasks such as text classification, text clustering, or information retrieval. FastText considers finer-grained character-level features, thus better handling non-vocabulary words and rare words. Furthermore, FastText has a small model size and fast training speed, making it suitable for processing large-scale text datasets.

[0078] In text classification tasks, the FastText model uses a multilayer perceptron (MLP) as the classifier, classifying text by learning the relationship between text representations and corresponding labels. By using the FastText model, efficient and accurate text classification can be achieved.

[0079] The city retrieval intent recognition method provided in this application uses a city dictionary, a special word dictionary, and a filtered city dictionary to make a preliminary determination of the city intent of the user's address retrieval information. At the same time, it combines a multi-label FastText model to predict the address retrieval information to obtain a sequence list. Based on the standard cities and the sequence list, it determines the target city to which the address retrieval information belongs, so as to retrieve the destination in the target city, avoid recalling too many redundant results, reduce the pressure of sorting, and the FastText model trained by the multi-label method improves the prediction accuracy of each city, narrows the gap between model training and prediction, and thus improves the accuracy of retrieval results.

[0080] In one embodiment, selecting target cities based on standard cities and the sequence list may specifically include:

[0081] Iterate through the first N candidate cities in the sequence list; where N is a positive integer;

[0082] Determine whether the candidate city encountered during the iteration is ranked first in the sequence list;

[0083] If so, add the candidate city to the city list;

[0084] If not, determine whether the distance between the candidate cities traversed and the user's city is less than the preset distance and whether the corresponding predicted probability is greater than or equal to the first preset threshold.

[0085] If so, add the candidate city to the first candidate list;

[0086] If not, determine whether the candidate cities encountered are the user's city;

[0087] If so, add the candidate city to the city list;

[0088] If not, determine whether the candidate cities encountered are standard cities;

[0089] If so, add the candidate city to the second candidate list;

[0090] The target city is selected based on the first candidate list, the second candidate list, and the city list.

[0091] This embodiment can iterate through the top 10 cities predicted by the FastText model, for example, by iterating in the order of the candidate cities in the sequence list. If a candidate city is the top 1 city predicted by the FastText model, then the city is directly added to the city list cityList.

[0092] Otherwise, determine whether the distance between the candidate city and the user's city is less than or equal to 200km and whether the corresponding predicted probability is greater than or equal to the first preset threshold, such as 0.5; if the conditions are met, add the candidate city to the first candidate list sortList.

[0093] Otherwise, further determine whether the candidate city is the user's city; if it is, directly add the candidate city to the city list cityList.

[0094] Otherwise, it determines whether the candidate cities encountered are standard cities, which are candidate cities with the administrative division suffix removed. For example, "city / county / district" etc., after removing "Shanghai", becomes "Shanghai". If it is a standard city, it is added to the second candidate list replaceList. Finally, the target city is selected based on the first candidate list, the second candidate list, and the city list to accurately determine the target city.

[0095] In one embodiment, selecting a target city based on the first candidate list, the second candidate list, and the city list includes:

[0096] Filter out the candidate cities that are closest to the user's city from the first candidate list and add them to the city list;

[0097] Add all candidate cities from the second candidate list to the city list;

[0098] Determine whether the number of candidate cities in the city list is greater than or equal to 3;

[0099] If so, select the target city from the city list.

[0100] In this embodiment, the first candidate list sortList can be sorted, and the candidate cities closest to the user's city can be extracted and added to the city list cityList. Then, it is determined whether the second candidate list replaceList is empty. If it is not empty, all candidate cities in the second candidate list are added to the city list cityList. If the number of cities in the city list cityList is greater than or equal to 3, the target cities are selected from the city list, such as selecting the best few cities as the return result. The number of target cities is less than or equal to 3.

[0101] In this embodiment, among the candidate cities predicted by the model, the model prediction probability of each candidate city and the distance between the candidate city and the user's city are comprehensively considered, and the best few cities are selected as the results to accurately determine the target city.

[0102] In one embodiment, after determining whether the number of candidate cities in the city list is greater than or equal to 3, the method further includes:

[0103] If not, iterate through the candidate cities in the sequence list from the Nth to the Mth position; where M is a positive integer greater than N.

[0104] Determine whether the candidate cities traversed are the closest to the user's city and have a predicted probability greater than a second preset threshold.

[0105] If so, add the candidate city to the city list.

[0106] If the number of candidate cities in the city list is less than 3, the top 50 cities predicted by the model are iterated through. If the current candidate city is city1 or city2 (the two cities closest to the user's city), and the model's prediction probability is >0.1 or the candidate city is ranked in the top 20, the candidate city is added to the city list cityList, and the top 50 cities are iterated through again until the number of candidate cities in the city list cityList is greater than or equal to 3, so as to accurately filter out the target city.

[0107] In one embodiment, before predicting the address retrieval information based on a multi-label FastText model and generating the sequence list, the method further includes:

[0108] Acquire training data; wherein the training data is fused data consisting of local data and cross-city data, and the fused data includes user click data, order data, POI data and address data;

[0109] The FastText model is trained using the training data in a multi-label manner to obtain a multi-label FastText model.

[0110] Specifically, this embodiment first acquires training data and performs necessary preprocessing, such as word segmentation, stop word removal, and stemming, to ensure that the training data is input into the model in a standardized form. Multi-label classification is then performed on the training data. Each sample should contain text content and multiple corresponding labels. Each label is encoded using an appropriate encoding method, such as one-hot encoding or a label encoder, to ensure that the labels can be correctly understood and processed during model training. A deep learning framework (such as PyTorch or TensorFlow) is used to build the model structure, and a loss function is defined for the multi-label classification task. This loss function includes binary cross-entropy loss, Jaccard loss, etc. The model is trained using the prepared training data and the defined loss function. Simultaneously, stochastic gradient descent (SGD) or other optimization algorithms are used to update the model parameters. The trained model is evaluated using a test set or cross-validation, and various metrics (such as accuracy, recall, F1 score, etc.) are calculated to assess model performance. When the evaluation results meet the requirements, a multi-label FastText model is obtained. Finally, the trained multi-label FastText model is used to classify text and output the corresponding labels.

[0111] In this embodiment, when training the FastText model, local and intercity data in the freight address scenario can be combined, covering user click data, order data, POI data, and address data as the training set of the model to enrich the data types of training. At the same time, a multi-label approach is used to train the FastText model to improve the prediction accuracy for each city, narrow the gap between model training and prediction, expand the value range of n-gram features, and enable the model to acquire more local information during the training process.

[0112] In one embodiment, after matching the address retrieval information with a city dictionary, the method further includes:

[0113] When a unique city is matched, that city is selected as the target city and recommended to the user.

[0114] This embodiment can extract the city name from the address search information and use the city dictionary lookup function to match the extracted city name in the dictionary. If the dictionary contains only the corresponding city name, the city name is returned. For example, if the user's address search information is "No. 188, Huaihai Middle Road, Shanghai", and the city dictionary matching strategy finds only one city, "Shanghai", then the result "Shanghai" is returned directly.

[0115] Reference Figure 2 This application also provides a city search intent recognition device, the device comprising:

[0116] Module 11 is used to obtain the user's address retrieval information;

[0117] The first matching module 12 is used to match the address retrieval information with a city dictionary;

[0118] The second matching module 13 is used to match the address retrieval information with a special word dictionary when the number of matched cities is greater than 1; wherein, the special word dictionary includes cities and corresponding special words representing the cities;

[0119] The third matching module 14 is used to match the address retrieval information using a filtered city dictionary when the number of matched cities is greater than 1, so as to obtain standard cities; wherein, the filtered city dictionary is a dictionary with the administrative division suffix removed.

[0120] Prediction module 15 is used to predict the address retrieval information based on the multi-label FastText model and generate a sequence list; wherein the sequence list includes multiple candidate cities and the predicted probability of each candidate city;

[0121] The filtering module 16 is used to filter target cities based on standard cities and the sequence list, and to recommend addresses belonging to the target cities to users.

[0122] As described above, it is understood that each component of the city search intent recognition device proposed in this application can realize the function of any of the city search intent recognition methods described above, and the specific structure will not be described in detail.

[0123] Reference Figure 3 This application also provides a computer device whose internal structure can be as follows: Figure 3 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes a storage medium and internal memory. The storage medium stores the operating system, computer programs, and the database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the storage medium. The database stores relevant data for the city search intent recognition method. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements a city search intent recognition method.

[0124] The processor described above executes the city search intent recognition method described above, the method comprising:

[0125] Obtain the user's address search information;

[0126] The address retrieval information is matched with the city dictionary;

[0127] When more than one city is matched, the address retrieval information is matched with a special word dictionary; wherein, the special word dictionary includes cities and corresponding special words representing the cities;

[0128] When more than one city is matched, the address retrieval information is matched using a filtered city dictionary to obtain standard cities; wherein, the filtered city dictionary is a dictionary with the administrative division suffix removed.

[0129] The address retrieval information is predicted based on the multi-label FastText model to generate a sequence list; wherein, the sequence list includes multiple candidate cities and the predicted probability of each candidate city;

[0130] Based on the standard cities and the sequence list, target cities are filtered out, and addresses belonging to the target cities are matched and recommended to the user.

[0131] One embodiment of this application also provides a computer-readable storage medium having a computer program stored thereon. When executed by a processor, the computer program implements a method for recognizing urban search intent, the method comprising:

[0132] Obtain the user's address search information;

[0133] The address retrieval information is matched with the city dictionary;

[0134] When more than one city is matched, the address retrieval information is matched with a special word dictionary; wherein, the special word dictionary includes cities and corresponding special words representing the cities;

[0135] When more than one city is matched, the address retrieval information is matched using a filtered city dictionary to obtain standard cities; wherein, the filtered city dictionary is a dictionary with the administrative division suffix removed.

[0136] The address retrieval information is predicted based on the multi-label FastText model to generate a sequence list; wherein, the sequence list includes multiple candidate cities and the predicted probability of each candidate city;

[0137] Based on the standard cities and the sequence list, target cities are filtered out, and addresses belonging to the target cities are matched and recommended to the user.

[0138] 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 computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media provided in this application and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0139] In summary, the greatest benefit of this application is that:

[0140] The present application provides a method, apparatus, device, and storage medium for identifying city search intent. It utilizes a city dictionary, a special word dictionary, and a filtered city dictionary to initially determine the city intent of a user's address search information. Simultaneously, it combines a multi-label FastText model to predict the address search information and obtain a sequence list. Based on standard cities and the sequence list, it determines the target city to which the address search information belongs, thus retrieving the destination within the target city. This avoids recalling too many redundant results, reduces the pressure of ranking, and the FastText model trained using a multi-label approach improves the prediction accuracy for each city, narrowing the gap between model training and prediction, thereby improving the accuracy of the search results.

[0141] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0142] The above description is only a preferred embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural changes made based on the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for identifying urban search intent, characterized in that, include: Obtain the user's address search information; The address retrieval information is matched with the city dictionary; When more than one city is matched, the address retrieval information is matched with a special word dictionary; wherein, the special word dictionary includes cities and corresponding special words representing the cities; When the address retrieval information is matched with a special word dictionary, and the number of matched cities is greater than 1, the address retrieval information is matched again using a filtered city dictionary to obtain standard cities; wherein, the filtered city dictionary is a dictionary with the administrative division suffix removed. The address retrieval information is predicted based on the multi-label FastText model to generate a sequence list; wherein, the sequence list includes multiple candidate cities and the predicted probability of each candidate city; Based on the standard cities and the sequence list, target cities are filtered out, and addresses belonging to the target cities are matched and recommended to the user.

2. The method according to claim 1, characterized in that, Before matching the address retrieval information with a special word dictionary, the process also includes: Obtain user click data and order data within a preset time period on the freight platform; Historical address retrieval information with unique matching cities is summarized from the click data and order data; The matching cities and corresponding special words representing the cities are filtered out from the historical address retrieval information to generate a special word dictionary.

3. The method according to claim 1, characterized in that, Target cities were selected based on standard cities and the aforementioned sequence list, including: Iterate through the first N candidate cities in the sequence list; where N is a positive integer; Determine whether the candidate city encountered during the iteration is ranked first in the sequence list; If so, add the candidate city to the city list; If not, determine whether the distance between the candidate cities traversed and the user's city is less than the preset distance and whether the corresponding predicted probability is greater than or equal to the first preset threshold. If the distance between the candidate city and the user's city is less than a preset distance and the corresponding predicted probability is greater than or equal to a first preset threshold, the candidate city is added to the first candidate list. If the distance between the candidate city and the user's city is less than the preset distance and the corresponding predicted probability is greater than or equal to the first preset threshold, determine whether the candidate city is the user's city. If the candidate city encountered during the iteration is the user's city, add the candidate city to the city list; If the candidate city encountered is not the user's city, determine whether the candidate city encountered is a standard city. If the candidate city encountered is a standard city, add the candidate city to the second candidate list; The target city is selected based on the first candidate list, the second candidate list, and the city list.

4. The method according to claim 3, characterized in that, Target cities were selected based on the first candidate list, the second candidate list, and the city list, including: Filter out the candidate cities that are closest to the user's city from the first candidate list and add them to the city list; Add all candidate cities from the second candidate list to the city list; Determine whether the number of candidate cities in the city list is greater than or equal to 3; If so, select the target city from the city list.

5. The method according to claim 4, characterized in that, After determining whether the number of candidate cities in the city list is greater than or equal to 3, the process also includes: If not, iterate through the candidate cities in the sequence list from the Nth to the Mth position; where M is a positive integer greater than N. Determine whether the candidate cities traversed are the closest to the user's city and have a predicted probability greater than a second preset threshold. If so, add the candidate city to the city list.

6. The method according to claim 1, characterized in that, Before generating the sequence list by predicting the address retrieval information based on the multi-label FastText model, the following steps are also included: Acquire training data; wherein the training data is fused data consisting of local data and cross-city data, and the fused data includes user click data, order data, POI data and address data; The FastText model is trained using the training data in a multi-label manner to obtain a multi-label FastText model.

7. The method according to claim 1, characterized in that, After matching the address retrieval information with the city dictionary, the process also includes: When a unique city is matched, that city is selected as the target city and recommended to the user.

8. A city search intent recognition device, characterized in that, include: The acquisition module is used to obtain the user's address retrieval information; The first matching module is used to match the address retrieval information with a city dictionary; The second matching module is used to match the address retrieval information with a special word dictionary when the number of matched cities is greater than 1; wherein, the special word dictionary includes cities and corresponding special words representing the cities; The third matching module is used to match the address retrieval information with a special word dictionary when the number of matched cities is greater than 1, and then use a filtered city dictionary to match the address retrieval information to obtain standard cities; wherein, the filtered city dictionary is a dictionary with the administrative division suffix removed. The prediction module is used to predict the address retrieval information based on the multi-label FastText model and generate a sequence list; wherein, the sequence list includes multiple candidate cities and the predicted probability of each candidate city; The filtering module is used to filter target cities based on standard cities and the sequence list, and to recommend addresses belonging to the target cities to users.

9. A computer device, characterized in that, include: processor; Memory; The memory stores a computer program, and when the processor executes the computer program, it implements the city search intent recognition method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the city search intent recognition method according to any one of claims 1-7.

Images