Point of interest recall method, device and equipment in freight scene and storage medium

Abstract

The application relates to a point-of-interest recall method and device in a freight scene, a computer device and a storage medium. The method comprises the following steps: receiving a position search request from a user terminal, wherein the position search request carries original query text and a city parameter; inputting the original query text and the city parameter into a multi-task omission model as input data, obtaining output data of the omission model, and the output data comprises omission probabilities corresponding to each character in the original query text; obtaining an omission query text according to the original query text and the output data; obtaining a first point-of-interest set according to the original query text and a second point-of-interest set according to the omission query text; and returning a search result to the user terminal according to the first point-of-interest set and the second point-of-interest set. The application adds a recall route, alleviates the problem of insufficient points of interest in the previous recall, enhances the search capability, enriches the search result, and enables the user to better select a site.

Description

Technical Field

[0001] This application relates to the field of address search technology, and in particular to a method, apparatus, computer equipment, and storage medium for recalling points of interest in a freight scenario. Background Technology

[0002] In freight business scenarios, when placing a freight order, users can search for the origin and / or destination by entering query text. After receiving the user's search request, the server will recall some points of interest based on the user's query text and provide them to the user, allowing the user to select a location.

[0003] Currently, when recalling points of interest, it is usually done directly based on the query text entered by the user. This method may result in an insufficient number of points of interest being recalled, which may prevent the user from selecting the address they want. Summary of the Invention

[0004] To address the aforementioned shortcomings or deficiencies, this application provides a method, apparatus, computer device, and storage medium for recalling points of interest in a freight scenario. The embodiments of this application add one recall channel, alleviating the problem of insufficient points of interest in previous recalls, enhancing search capabilities, enriching search results, and enabling users to select locations more effectively.

[0005] This application provides a method for recalling points of interest in a freight scenario according to a first aspect. In one embodiment, the method includes:

[0006] Receive location search requests from user terminals. The location search requests carry the original query text and city parameters.

[0007] The original query text and city parameters are input into a pre-trained omission model. The output data of the omission model is obtained, which includes the omission probability of each character in the original query text.

[0008] The omitted query text is obtained based on the original query text and the output data;

[0009] The first set of points of interest is retrieved based on the original query text, and the second set of points of interest is retrieved based on the omitted query text.

[0010] Search results are returned to the user's terminal based on the first set of points of interest and the second set of points of interest.

[0011] In some embodiments, the model training process is omitted, and includes:

[0012] Retrieve multiple historical user order logs for freight operations; each historical user order log includes historical query text and historical city parameters;

[0013] Multiple training data sets were constructed based on the above-mentioned historical user order logs; each training data set includes historical city parameters, historical query text, and the corresponding annotation information of the historical query text.

[0014] Based on the above-mentioned multiple historical user order logs, the city click distribution information corresponding to various historical query texts is obtained; the city click distribution information corresponding to any historical query text includes the distribution probability of each order city corresponding to that historical query text.

[0015] Construct the initial omission model;

[0016] Based on the above training data and the city click distribution information corresponding to various historical query texts, the initial omission model is iteratively trained until the preset training termination condition is met, and a trained omission model is obtained.

[0017] In some embodiments, the initial omission model is iteratively trained based on the aforementioned multiple training data and the city click distribution information corresponding to various historical query texts, including:

[0018] During each training session, the training data used in this session is used as the target training data. The historical city parameters and historical query text in the target training data are used as input data to the initial omission model. The output data of the initial omission model is obtained. The output data includes city distribution prediction information and omission mark prediction information. The city distribution prediction information includes the predicted distribution probability of each order-issuing city corresponding to the historical query text in the target training data. The omission mark prediction information includes the omission probability corresponding to each character in the historical query text in the target training data.

[0019] Obtain the city click distribution information and annotation information corresponding to the historical query text in the target training data as the target distribution information and target annotation information;

[0020] The first loss is calculated based on the city distribution prediction information and the target distribution information;

[0021] The second loss is calculated based on the omission marker prediction information and the target annotation information;

[0022] Determine whether the preset training termination condition is met based on the first loss and the second loss;

[0023] If the condition is met, training stops; if not, the next training session begins.

[0024] In some embodiments, calculating a first loss based on city distribution prediction information and target distribution information includes:

[0025] The first loss is calculated based on the following formula;

[0026]

[0027] Among them, L city This is the first loss, where n is the total number of cities. It is the distribution probability of the i-th city in the target distribution information. It is the predicted distribution probability of the i-th city that issues the order in the city distribution prediction information;

[0028] The second loss is calculated based on the omission marker prediction information and the target annotation information, including:

[0029] The second loss is calculated based on the following formula;

[0030]

[0031] Among them, L omit This is the second loss, with α set to 0.05. This is the omission of the predicted information.

[0032] In some embodiments, each historical user order log also includes the point of interest name and point of interest address;

[0033] Based on the above historical user order logs, multiple training data sets were constructed, including:

[0034] The annotation program automatically annotates the historical query text in each historical user order log, and uses each historical query text in the historical user order log and its corresponding annotation data as a training data point to obtain multiple training data points.

[0035] The operations for annotating the historical query text in each historical user order log include:

[0036] Retrieve the historical query text, point of interest name, and point of interest address for each historical user order log;

[0037] Determine whether each character of the historical query text appears in the point of interest name or point of interest address;

[0038] If so, mark the character as 0;

[0039] If not, mark the character as 1.

[0040] In some embodiments, obtaining the omitted query text based on the original query text and output data includes:

[0041] Delete the target character from the original query text to obtain the omitted query text; the target character is the character whose omission probability is higher than the preset probability value.

[0042] In some embodiments, returning search results to a user terminal based on a first set of points of interest and a second set of points of interest includes:

[0043] Detect whether there are any recall results in the first set of points of interest;

[0044] In response to a positive detection result, the first set of points of interest is returned to the user terminal as the search result.

[0045] If the detection result is negative, both the first set of points of interest and the second set of points of interest will be returned to the user terminal as search results.

[0046] This application provides a point-of-interest recall device for freight scenarios according to a second aspect. In one embodiment, the device includes:

[0047] The receiving module is used to receive location search requests from user terminals. The location search requests carry the original query text and city parameters.

[0048] The prediction module is used to input the original query text and city parameters as input data into a pre-trained omission model and obtain the output data of the omission model. The output data includes the omission probability of each character in the original query text.

[0049] The omitted text acquisition module is used to obtain the omitted query text based on the original query text and the output data.

[0050] The recall module is used to recall the first set of points of interest based on the original query text, and to recall the second set of points of interest based on the omitted query text;

[0051] The results return module is used to return search results to the user terminal based on the first set of points of interest and the second set of points of interest.

[0052] This application provides a computer device according to a third aspect, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any of the above-described methods.

[0053] According to a fourth aspect, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-described methods.

[0054] In this embodiment, the server receives a location search request from a user terminal, the request carrying original query text and city parameters; the original query text and city parameters are input into a pre-trained omission model, and the output data of the omission model is obtained, including the omission probability corresponding to each character in the original query text; the omitted query text is obtained based on the original query text and the output data; a first set of points of interest is retrieved based on the original query text, and a second set of points of interest is retrieved based on the omitted query text; the search results are returned to the user terminal based on the first set of points of interest and the second set of points of interest. This embodiment, based on the use of the original query text for point of interest retrieval, adds another retrieval path, alleviating the problem of insufficient points of interest in previous retrieval methods, enhancing search capabilities, enriching search results, and enabling users to select locations more effectively. Attached Figure Description

[0055] Figure 1 This is a flowchart illustrating a point-of-interest (POI) retrieval method in a freight scenario, as shown in one embodiment.

[0056] Figure 2 This is a flowchart illustrating the training process of an omission model in one embodiment;

[0057] Figure 3 This is a schematic diagram of the model structure of the initial omitted model provided in one embodiment;

[0058] Figure 4 This is a structural block diagram of a point-of-interest recall device in a freight scenario, as shown in one embodiment.

[0059] Figure 5 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0060] 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.

[0061] This application provides a method for recalling points of interest in a freight transportation scenario. In some embodiments, the method includes, for example... Figure 1 The steps shown below will be explained using the method applied to a server as an example.

[0062] S110: Receives a location search request from a user terminal. The location search request carries the original query text and city parameters.

[0063] The method provided in this application can be used in scenarios where users search for freight addresses in real time. Users can send a location search request to the server through a user terminal. This request carries the original query text entered by the user (including at least one character) and city parameters (specifically, the city code or the city name). The user terminal is a computing device used by the user, which can be a smart mobile device such as a smartphone or tablet.

[0064] The query text in this application is usually referred to as a query in English, and can also be translated as query.

[0065] S120: Input the original query text and city parameters into the pre-trained omission model, and obtain the output data of the omission model. The output data includes the omission probability of each character in the original query text.

[0066] This step uses an ellipsis model to determine whether individual characters in the original query text need to be omitted. This ellipsis model is pre-trained offline, and the training process will be described in detail below.

[0067] S130: Obtain the omitted query text based on the original query text and the output data.

[0068] The output data provides the omission probability for each character in the original query text. Therefore, when obtaining the omitted query text based on the original query text and the output data, the target character in the original query text can be deleted to obtain the omitted query text. The target character is the character whose omission probability is higher than a preset probability value. This preset probability value can be flexibly set and adjusted according to actual conditions; this embodiment does not impose any restrictions.

[0069] S140: Retrieve the first set of points of interest based on the original query text, and retrieve the second set of points of interest based on the omitted query text.

[0070] This step retrieves points of interest using both the original query text and the omitted query text. The points of interest retrieved based on the original query text constitute the first set of points of interest, while the points of interest retrieved based on the omitted query text constitute the second set of points of interest.

[0071] The operation of recalling points of interest using query text (such as the original query text or the omitted query text) is irrelevant to the improvements of this application and therefore will not be described in detail. Those skilled in the art will understand the specific implementation details of this operation.

[0072] S150: Return search results to the user terminal based on the first set of points of interest and the second set of points of interest.

[0073] In one implementation, the first set of points of interest and the second set of points of interest can be merged, and the merged set can be used as the search result and returned to the user terminal.

[0074] In another embodiment, returning search results to a user terminal based on a first set of points of interest and a second set of points of interest includes: detecting whether there are recall results for the first set of points of interest; in response to a yes detection result, returning the first set of points of interest as a search result to the user terminal; and in response to a no detection result, returning both the first set of points of interest and the second set of points of interest as search results to the user terminal.

[0075] This implementation sets priorities for the original query text and the omitted query text. If a recall result can be obtained through the original query text, the points of interest recalled through the omitted query text will not be used. Conversely, if a recall result cannot be obtained through the original query text, the first set of points of interest and the second set of points of interest will be merged, and the merged set will be used as the search result and returned to the user terminal.

[0076] The above embodiments of this application, based on the use of original query text for point of interest retrieval, add an additional retrieval path, alleviating the problem of insufficient points of interest in previous retrieval, enhancing search capabilities, enriching search results, and enabling users to select locations more effectively.

[0077] In some embodiments, the model training process is omitted, such as... Figure 2 As shown, it includes:

[0078] S210: Obtain multiple historical user order logs for freight business; each historical user order log includes historical query text and historical city parameters.

[0079] User order logs record important information such as the query text and city parameters corresponding to the freight order when the user initiates it. Historical user order logs refer to user order logs whose operation time is before the current time.

[0080] In some implementations, user click logs can also be used to construct training data, as these logs record the query text clicked by users. However, considering that user clicks may be biased, such as accidental clicks during a search, which could interfere with model training, this embodiment prefers to use user click logs to construct training data to achieve better model prediction results. The historical query text refers to query text whose corresponding time is before the current time. Similarly, historical city parameters and the city parameters currently entered by the user are both city parameters, except that the latter corresponds to the current time, while the former corresponds to a time before the current time.

[0081] S220: Construct multiple pieces of training data based on the above-mentioned multiple historical user order-issuing logs; each piece of training data includes historical city parameters, a historical query text, and the annotation information corresponding to the historical query text.

[0082] Specifically, each historical user order-issuing log further includes a point of interest name and a point of interest address. The point of interest name and the point of interest address specifically refer to the name and address of the point of interest for which the order is issued. Correspondingly, when constructing multiple pieces of training data based on the above-mentioned multiple historical user order-issuing logs, manual annotation is not required. Instead, an annotation program can automatically annotate the historical query text in each historical user order-issuing log, and use the historical query text in each historical user order-issuing log and its corresponding annotation data as a piece of training data to obtain multiple pieces of training data. Automatically annotating through a program can significantly improve the annotation efficiency, and thus improve the training efficiency of the model.

[0083] Among them, the operation of annotating the historical query text in each historical user order-issuing log includes: obtaining the historical query text, the point of interest name, and the point of interest address included in each historical user order-issuing log; determining whether each character of the historical query text appears in the point of interest name or the point of interest address; if so, marking the character as 0; if not, marking the character as 1.

[0084] Exemplarily, assume that the historical query text is "Fourth floor of Hankou North Commodity City, phone 130000", the point of interest name for which the order is issued is "Commodity City", and the point of interest address is "Wuhan - Huangpi District - Hankou North Avenue Hankou North International Trade Center". First, the historical query text can be character-sliced, and the slicing result is ['汉', '囗', '北', '日', '用', '品', '商', '贸', '城', '四', '楼', ',', '电', '话', '130000']. Among them, the characters '商', '贸', '电', '话', '130000' do not appear in the point of interest name or the point of interest address, so they are marked as 1, and the remaining characters are marked as 0. Finally, the marking situation of each character in the historical query text is as follows:

[0085] "汉 / 0囗 / 0北 / 0日 / 0用 / 0品 / 0商 / 1贸 / 1城 / 0四 / 1楼 / 1, / 1电 / 1话 / 1 130000 / 1".

[0086] S230: Perform aggregation statistics based on the above-mentioned multiple historical user order-issuing logs to obtain the city click distribution information corresponding to various historical query texts; the city click distribution information corresponding to any historical query text includes the distribution probability of each order-issuing city corresponding to the historical query text.

[0087] Considering that users may include city parameters when searching for origin and / or destination in freight scenarios, and that users may intend to travel across cities (e.g., originating in "Suzhou" but searching for a destination in "Shanghai"), this embodiment also labels the cities. Specifically, it statistically analyzes the probability distribution of different order-issuing cities under the same query text, using this as the ground truth for multi-labeling. This allows the model to learn different omission patterns of the same query text across different cities, improving the accuracy of the model's prediction of omitted query text. Furthermore, identical historical query texts from different users' order-issuing logs are considered the same historical query text.

[0088] For example, suppose a certain historical query text is "China World Trade Center", and the cities that issued the order for this historical query text are "Beijing", "Shanghai", "Shenzhen" and "Chengdu", with the number of orders issued for each city being "10", "9", "8" and "7" respectively. Then the probability distribution of each city can be calculated as [0.2941, 0.2647, 0.2353, 0.2059].

[0089] S240: Construct the initial omission model.

[0090] The initial omission model can be a commonly used deep learning model. For example, the structure of the initial omission model can be found here. Figure 3 The model's input consists of query text and city parameters. This input is first encoded by an encoder, and then the processed data is further processed by Token Dense and CLS (CLS stands for classification) Dense layers to obtain the first and second losses. The Dense layer (fully connected layer) is a commonly used layer in deep neural networks, helping the neural network better understand the input data and thus improving the model's accuracy.

[0091] S250: Based on the above training data and the city click distribution information corresponding to various historical query texts, the initial omission model is iteratively trained until the preset training termination condition is met, and the trained omission model is obtained.

[0092] In each training session, the training data used in this session is used as the target training data. The historical city parameters and historical query text in the target training data are used as input data to the initial omission model. The output data of the initial omission model is obtained. The output data includes city distribution prediction information and omission mark prediction information. The city distribution prediction information includes the predicted distribution probability of each order-issuing city corresponding to the historical query text in the target training data. The omission mark prediction information includes the omission probability corresponding to each character in the historical query text in the target training data.

[0093] Obtain the city click distribution information and annotation information corresponding to the historical query text in the target training data as the target distribution information and target annotation information;

[0094] The first loss is calculated based on the city distribution prediction information and the target distribution information;

[0095] The second loss is calculated based on the omission marker prediction information and the target annotation information;

[0096] Determine whether the preset training termination condition is met based on the first loss and the second loss;

[0097] If the condition is met, training stops; if not, the next training session begins.

[0098] Specifically, the server can calculate the first loss based on the following formula;

[0099]

[0100] Among them, L city This is the first loss, where n is the total number of cities. It is the distribution probability of the i-th city in the target distribution information. It is the predicted distribution probability of the i-th city that issues the order in the city distribution prediction information.

[0101] Accordingly, the server also calculates the second loss based on the following formula;

[0102]

[0103] Among them, L omit This is the second loss, with α set to 0.05. This is the omission of the predicted information.

[0104] This embodiment focuses on optimizing the loss. The initial omission model has two tasks: the first is to predict the distribution probability of cities, which yields city distribution prediction information; the second is a binary classification task (determining whether a character needs to be omitted), which yields omission marker prediction information. The loss for the first task is a multi-label loss, i.e., the first loss, and the loss for the second task is the second loss. For the second task, label smoothing can be added; that is, if the labeled value is 1, the corresponding loss uses 1-α, where α can be set to 0.05.

[0105] After calculating the first loss and the second loss, they are summed, and then the total loss obtained from the sum is used to determine whether to end the training.

[0106] In some embodiments, the TensorFlow deep learning framework (or alternatively, Torch, Keras, etc.) can be used during model training. A v100 graphics card is used as the GPU (graphics processing unit). In the model parameter settings, the batch size can be set to 512. The AdamW optimizer is also used. An early stopping strategy is set for the loss (referring to the total loss mentioned above). If the loss does not decrease after 10 iterations, the training is terminated.

[0107] After training is complete, the initial omission model can be saved to a relevant server, such as Alibaba Cloud's OSS server. The server can then pull the trained omission model from the relevant server and load the model for subsequent online real-time prediction.

[0108] Figure 1-2 This is a flowchart illustrating a point-of-interest (POI) retrieval method in a freight scenario, as shown in one embodiment. It should be understood that, although... Figure 1-2 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1-2 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0109] Based on the same inventive concept, this application also provides a point-of-interest recall device for freight scenarios. In some embodiments, such as Figure 4 As shown, the device includes the following modules:

[0110] The receiving module 110 is used to receive a location search request from a user terminal, the location search request carrying the original query text and city parameters;

[0111] Prediction module 120 is used to input the original query text and city parameters as input data into a pre-trained omission model and obtain the output data of the omission model. The output data includes the omission probability corresponding to each character in the original query text.

[0112] The omitted text acquisition module 130 is used to obtain the omitted query text based on the original query text and the output data;

[0113] The recall module 140 is used to recall a first set of points of interest based on the original query text, and to recall a second set of points of interest based on the omitted query text.

[0114] The result return module 150 is used to return search results to the user terminal based on the first set of points of interest and the second set of points of interest.

[0115] In some embodiments, the apparatus further includes a training module for training the omission model.

[0116] The training module trains the omitted model, including:

[0117] Retrieve multiple historical user order logs for freight operations; each historical user order log includes historical query text and historical city parameters;

[0118] Multiple training data sets were constructed based on the above-mentioned historical user order logs; each training data set includes historical city parameters, historical query text, and the corresponding annotation information of the historical query text.

[0119] Based on the above-mentioned multiple historical user order logs, the city click distribution information corresponding to various historical query texts is obtained; the city click distribution information corresponding to any historical query text includes the distribution probability of each order city corresponding to that historical query text.

[0120] Construct the initial omission model;

[0121] Based on the above training data and the city click distribution information corresponding to various historical query texts, the initial omission model is iteratively trained until the preset training termination condition is met, and a trained omission model is obtained.

[0122] In some embodiments, when the training module iteratively trains the initial omission model based on the aforementioned multiple training data and the city click distribution information corresponding to various historical query texts, it is used for:

[0123] During each training session, the training data used in this session is used as the target training data. The historical city parameters and historical query text in the target training data are used as input data to the initial omission model. The output data of the initial omission model is obtained. The output data includes city distribution prediction information and omission mark prediction information. The city distribution prediction information includes the predicted distribution probability of each order-issuing city corresponding to the historical query text in the target training data. The omission mark prediction information includes the omission probability corresponding to each character in the historical query text in the target training data.

[0124] Obtain the city click distribution information and annotation information corresponding to the historical query text in the target training data as the target distribution information and target annotation information;

[0125] The first loss is calculated based on the city distribution prediction information and the target distribution information;

[0126] The second loss is calculated based on the omission marker prediction information and the target annotation information;

[0127] Determine whether the preset training termination condition is met based on the first loss and the second loss;

[0128] If the condition is met, training stops; if not, the next training session begins.

[0129] In some embodiments, the training module calculates the first loss based on the following formula;

[0130]

[0131] Among them, L city This is the first loss, where n is the total number of cities. It is the distribution probability of the i-th city in the target distribution information. It is the predicted distribution probability of the i-th city that issues the order in the city distribution prediction information;

[0132] The training module also calculates the second loss based on the following formula;

[0133]

[0134] Among them, L omit This is the second loss, with α set to 0.05. This is the omission of the predicted information.

[0135] In some embodiments, each historical user order log also includes an interest point name and an interest point address; when the training module constructs multiple training data based on the above multiple historical user order logs, it is used for:

[0136] The annotation program automatically annotates the historical query text in each historical user order log, and uses each historical query text in the historical user order log and its corresponding annotation data as a training data point to obtain multiple training data points.

[0137] The operations for annotating the historical query text in each historical user order log include:

[0138] Retrieve the historical query text, point of interest name, and point of interest address for each historical user order log;

[0139] Determine whether each character of the historical query text appears in the point of interest name or point of interest address;

[0140] If so, mark the character as 0;

[0141] If not, mark the character as 1.

[0142] In some embodiments, the omitted text acquisition module 130 is specifically used for:

[0143] Delete the target character from the original query text to obtain the omitted query text; the target character is the character whose omission probability is higher than the preset probability value.

[0144] In some embodiments, the result return module 150 is specifically used for:

[0145] Detect whether there are any recall results in the first set of points of interest;

[0146] In response to a positive detection result, the first set of points of interest is returned to the user terminal as the search result.

[0147] If the detection result is negative, both the first set of points of interest and the second set of points of interest will be returned to the user terminal as search results.

[0148] Specific limitations regarding the point-of-interest (POI) recall device in freight scenarios can be found in the limitations of the POI recall method in freight scenarios described above, and will not be repeated here. Each module in the aforementioned POI recall device for freight scenarios can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0149] In one embodiment, a computer device is provided, the internal structure of which can be shown as follows: Figure 5 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile 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 non-volatile storage medium. The database stores training data and other data; the specific data stored can be further defined in the above method embodiments. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a point-of-interest (POI) retrieval method in a freight transportation scenario.

[0150] Those skilled in the art will understand that Figure 5The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0151] This embodiment also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the method provided in any of the above method embodiments.

[0152] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the method provided in any of the above method embodiments.

[0153] Those skilled in the art will understand that implementing all or part of the processes in the above method embodiments can be accomplished by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application 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 data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0154] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0155] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for recalling points of interest in a freight transportation scenario, characterized in that, The method includes: Receive a location search request from a user terminal, the location search request carrying the original query text and city parameters; The original query text and the city parameters are input into a pre-trained omission model to obtain the output data of the omission model. The output data includes the omission probability of each character in the original query text. The omitted query text is obtained based on the original query text and the output data; A first set of points of interest is obtained based on the original query text, and a second set of points of interest is obtained based on the omitted query text; Search results are returned to the user terminal based on the first set of points of interest and the second set of points of interest. The training process of the omitted model includes: Obtain multiple historical user order logs for freight operations; each historical user order log includes historical query text and historical city parameters. Multiple training data sets are constructed based on the multiple historical user order logs; each training data set includes historical city parameters, historical query text, and annotation information corresponding to the historical query text. Based on the aggregated statistics of the multiple historical user order logs, the city click distribution information corresponding to various historical query texts is obtained; the city click distribution information corresponding to any historical query text includes the distribution probability of each order city corresponding to that historical query text. Construct the initial omission model; The initial omission model is iteratively trained based on the multiple training data and the city click distribution information corresponding to the various historical query texts until the preset training termination condition is met, thus obtaining the trained omission model.

2. The method as described in claim 1, characterized in that, The initial omission model is iteratively trained based on the multiple training data and the city click distribution information corresponding to the various historical query texts, including: During each training session, the training data used in this session is used as the target training data. The historical city parameters and historical query text in the target training data are used as input data to the initial omission model. The output data of the initial omission model is obtained. The output data includes city distribution prediction information and omission marker prediction information. The city distribution prediction information includes the predicted distribution probability of each order-issuing city corresponding to the historical query text in the target training data. The omission marker prediction information includes the omission probability corresponding to each character in the historical query text in the target training data. Obtain the city click distribution information and annotation information corresponding to the historical query text in the target training data as the target distribution information and target annotation information; Calculate the first loss based on the city distribution prediction information and the target distribution information; Calculate the second loss based on the omission marker prediction information and the target annotation information; Determine whether the preset training termination condition is met based on the first loss and the second loss; If the condition is met, training stops; if not, the next training session begins.

3. The method as described in claim 2, characterized in that, The first loss is calculated based on the city distribution prediction information and the target distribution information, including: The first loss is calculated based on the following formula; in, This is the first loss, where n is the total number of cities. It is the distribution probability of the i-th city that issues the order in the target distribution information. It is the predicted distribution probability of the i-th city that issues the order in the city distribution prediction information; The second loss is calculated based on the omission marker prediction information and the target annotation information, including: The second loss is calculated based on the following formula; in, This is the second loss, with α set to 0.

05. This is the omission marker prediction information.

4. The method as described in claim 2, characterized in that, Each of the aforementioned historical user order logs also includes the point of interest name and point of interest address; Multiple training data sets were constructed based on the aforementioned historical user order logs, including: The annotation program automatically annotates the historical query text in each of the historical user order logs, and uses each historical query text in the historical user order log and its corresponding annotation data as a training data to obtain multiple training data. The operation of annotating the historical query text in each of the aforementioned historical user order logs includes: Obtain the historical query text, point of interest name, and point of interest address included in each of the aforementioned historical user order logs; Determine whether each character of the historical query text appears in the point of interest name or point of interest address; If so, mark the character as 0; If not, mark the character as 1.

5. The method as described in claim 1, characterized in that, The omitted query text is obtained based on the original query text and the output data, including: Delete the target character from the original query text to obtain the omitted query text; the target character is a character whose omission probability is higher than a preset probability value.

6. The method as described in claim 1, characterized in that, Search results are returned to the user terminal based on the first set of points of interest and the second set of points of interest, including: Check if there are any recall results in the first set of points of interest; In response to a positive detection result, the first set of points of interest is returned as a search result to the user terminal. If the detection result is negative, both the first set of points of interest and the second set of points of interest are returned as search results to the user terminal.

7. A point-of-interest recall device for freight scenarios, characterized in that, The device includes: The receiving module is used to receive location search requests from user terminals, the location search requests carrying original query text and city parameters; The prediction module is used to input the original query text and the city parameters as input data into a pre-trained omission model, and obtain the output data of the omission model. The output data includes the omission probability corresponding to each character in the original query text. An omitted text acquisition module is used to obtain the omitted query text based on the original query text and the output data; The recall module is used to recall a first set of points of interest based on the original query text, and to recall a second set of points of interest based on the omitted query text. The result return module is used to return search results to the user terminal based on the first set of points of interest and the second set of points of interest. The training process of the omitted model includes: Obtain multiple historical user order logs for freight operations; each historical user order log includes historical query text and historical city parameters. Multiple training data sets are constructed based on the multiple historical user order logs; each training data set includes historical city parameters, historical query text, and annotation information corresponding to the historical query text. Based on the aggregated statistics of the multiple historical user order logs, the city click distribution information corresponding to various historical query texts is obtained; the city click distribution information corresponding to any historical query text includes the distribution probability of each order city corresponding to that historical query text. Construct the initial omission model; The initial omission model is iteratively trained based on the multiple training data and the city click distribution information corresponding to the various historical query texts until the preset training termination condition is met, thus obtaining the trained omission model.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images