POI search model training method and device
By training the POI search model, using the text similarity between the search term samples and POI text samples, the problems of inaccurate and incomplete POI search results in the prior art are solved, and more accurate search results are achieved.
Patent Information
- Application Number
- CN202411898472.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-20
- Publication Date
- 2025-03-28
AI Technical Summary
The POI search results in the prior art are inaccurate and incomplete, mainly because traditional models cannot fully understand the semantic information of the text, and it is difficult to accurately capture the ambiguity and ambiguity of search keywords.
A training method for POI search model is provided. By determining the training strategy data, the search term sample and POI text sample are obtained, and the text similarity is calculated. Based on these data, the model is trained to generate a model that can match the POI text according to the input search term output.
By combining the characteristics of the POI search scenario, the accurate matching of the search terms and output POI text is achieved, and the search intention is accurately identified, thereby improving the accuracy and comprehensiveness of the POI search results.
Smart Images
Figure CN119719306A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of POI search, and more particularly to a method for training a POI search model. The present application also relates to a training device for a POI search model, an electronic device, and a computer-readable storage medium. Background Art
[0002] For POI search scenarios, most traditional search engines use probabilistic models such as BM25 as the mainstream search solution. The BM25 model is based on a statistical method based on word frequency and cannot fully understand the semantic information of the text. Therefore, it performs relatively poorly when dealing with complex semantic relationships. In actual applications, the search keywords may be polysemous and ambiguous, and traditional models are difficult to accurately capture the semantic meaning of the text, resulting in inaccurate and incomplete POI search results. Summary of the invention
[0003] The present invention provides a training method, apparatus, terminal device and computer-readable storage medium for a POI search model to solve the problem of inaccurate and incomplete POI search results in the prior art.
[0004] In order to solve or improve the above technical problems to a certain extent, according to one aspect of the present invention, a training method for a POI search model is provided, comprising:
[0005] Determine the training strategy data corresponding to the POI search model to be trained;
[0006] Obtaining a search word sample and a POI text sample corresponding to the training strategy data, and obtaining a text similarity between the search word sample and the POI text sample;
[0007] Model training is performed based on the search word samples, the POI text samples and the text similarity to obtain the POI search model, which is used to output a target POI text that matches the training strategy data according to an input target search word.
[0008] In some implementations, determining the training strategy data corresponding to the POI search model to be trained includes:
[0009] Obtain users' real search demand data for specific search scenarios;
[0010] Based on the actual search demand data, training strategy data corresponding to the POI search model to be trained is determined.
[0011] In some embodiments, the specific search scenario includes: a POI search scenario in which a first search term includes a popular address name, the popular address name is inconsistent with the address name of a first POI targeted by the first search term, and the geographical location of the popular address name is within a predetermined distance range of the first POI;
[0012] The actual search demand data includes: searching for the first POI based on the first search term;
[0013] The determining of the training strategy data corresponding to the POI search model to be trained includes: performing text convergence model training based on the popular address name and the address name of the first POI, so that the trained POI search model can output the first POI based on the first search term; wherein the popular address name represents an address name for which the user's attention exceeds a predetermined attention threshold;
[0014] The obtaining of the search term sample and the POI text sample corresponding to the training strategy data includes: taking the first search term containing the popular address name as the search term sample, and taking the first POI as the POI text sample.
[0015] In some implementations, the specific search scenario includes: a POI search scenario for a second search term that includes a geographical term;
[0016] The real search demand data includes: searching, based on the second search term, for POIs whose relevance to the geographical term exceeds a predetermined relevance threshold;
[0017] The determining of the training strategy data corresponding to the POI search model to be trained includes: performing model training based on the association between the geographical term and the second POI corresponding to the second search term, so that the trained POI search model can output a POI whose association with the geographical term exceeds a predetermined association threshold based on the second search term;
[0018] The obtaining of the search term sample and the POI text sample corresponding to the training strategy data includes: taking the second search term containing the geographical term as the search term sample, and taking the second POI as the POI text sample.
[0019] In some implementations, the specific search scenario includes: a POI search scenario in which a third search term and its corresponding third POI are synonyms or homophones;
[0020] The actual search demand data includes: searching for the third POI based on the third search term;
[0021] The determining of the training strategy data corresponding to the POI search model to be trained includes: performing text convergence model training based on the third search term and the third POI, so that the trained POI search model can output the third POI based on the third search term;
[0022] The obtaining of the search term sample and the POI text sample corresponding to the training strategy data includes: taking the third search term as the search term sample and taking the third POI as the POI text sample.
[0023] In some implementations, obtaining the text similarity between the search term sample and the POI text sample according to the training strategy data includes one of the following:
[0024] Calculating the text similarity between the search term sample and the POI text sample based on a preset text similarity calculation method;
[0025] Scoring the search term samples and the POI text samples according to the correlation between the search term and the POI text represented by the training strategy data to obtain the text similarity between the search term samples and the POI text samples;
[0026] An initial similarity score between the search term sample and the POI text sample is calculated based on a preset text similarity calculation method, and the initial similarity score is adjusted according to the correlation between the search term and the POI text represented by the training strategy data to obtain the text similarity.
[0027] In some embodiments, the determining the training strategy data corresponding to the POI search model to be trained includes: determining the degree of matching between the search term data and the POI recommendation data according to the search term data input by the user in the historical POI search scene, the POI recommendation data output to the user, and the selection result of the user for the POI recommendation data, and determining the training strategy data corresponding to the POI search model to be trained based on the matching degree;
[0028] The obtaining of the search term samples and POI text samples corresponding to the training strategy data includes: in response to a matching degree between the search term data and the POI recommendation data exceeding a predetermined matching threshold, taking the search term data as the search term sample and taking the POI recommendation data as the POI text sample.
[0029] According to another aspect of the present invention, a training device for a POI search model is provided, the device comprising:
[0030] A training strategy data determining unit, used to determine the training strategy data corresponding to the POI search model to be trained;
[0031] A training sample obtaining unit, used to obtain a search word sample and a POI text sample corresponding to the training strategy data, and obtain a text similarity between the search word sample and the POI text sample;
[0032] A model training unit is used to perform model training based on the search word sample, the POI text sample and the text similarity to obtain the POI search model, wherein the POI search model is used to output a target POI text matching the training strategy data according to an input target search word.
[0033] According to another aspect of the present invention, there is provided a terminal device, comprising a processor and a memory; wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement the above method.
[0034] According to another aspect of the present invention, a computer-readable storage medium is provided, on which one or more computer instructions are stored, and the instructions are executed by a processor to implement the above method.
[0035] Compared with the prior art, the present invention has the following advantages:
[0036] The training method of the POI search model provided by the present invention includes: determining the training strategy data corresponding to the POI search model to be trained; obtaining the search word samples and POI text samples corresponding to the training strategy data, and obtaining the text similarity between the search word samples and the POI text samples; performing model training based on the search word samples, the POI text samples and the text similarity to obtain the POI search model, which is used to output the target POI text matching the training strategy data according to the input target search word. The POI search model obtained by the training of the present invention can be combined with the characteristics of the POI search scene to achieve accurate matching of the search word and the output POI text, so as to accurately identify the search intent of the POI search scene, thereby making the POI search results more accurate. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 is a flow chart of a training method for a POI search model provided in an embodiment of the present application;
[0038] Figure 2 It is a unit block diagram of a training device for a POI search model provided in an embodiment of the present application;
[0039] Figure 3 is a schematic diagram of the logical structure of an electronic device provided in an embodiment of the present application; DETAILED DESCRIPTION
[0040] Many specific details are described in the following description to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of the present application, so the present application is not limited by the specific implementation disclosed below.
[0041] For POI search scenarios, in order to improve the accuracy and comprehensiveness of POI search results, the present application provides a POI search model training method, a POI search model training device corresponding to the method, an electronic device, and a computer-readable storage medium. The following provides embodiments to describe the above method, device, electronic device, and computer-readable storage medium in detail.
[0042] The first embodiment of the present application provides a method for training a POI search model, and the application subject of the method can be a computing device application for model training. Figure 1 This is a flowchart of the training method of the POI search model provided in the first embodiment of the present application. Figure 1 The method provided in this embodiment is described in detail. The embodiments described below are used to explain the principle of the method and are not intended to be limiting for actual use.
[0043] like Figure 1 As shown, the training method of the POI search model provided in this embodiment includes the following steps:
[0044] S101, determining training strategy data corresponding to a POI search model to be trained.
[0045] POI (Point of Interest) search is an important geographic information service. In actual use, users initiate searches by entering keywords, selecting categories, or using markers on the map. After receiving the user's request, the system will match the input information in the database. Based on the matching results, the system will display the most relevant POI information to the user. The user can select one or more points of interest and obtain routes to these locations.
[0046] The POI search model based on text similarity mainly relies on text analysis and processing technology to understand the correlation between user queries and POI information. In the POI search process, different search subjects have different search needs for different search scenarios. Different search behavior habits correspond to the differences in the knowledge reserves, language habits, and environmental cognition of the search subjects, so that the search needs of different search subjects are specific and differentiated, which is mainly reflected in the high ambiguity and uncertainty of the search terms entered by the search subjects. In order to enable the trained POI search model to approach the POI search needs covering all scenarios, for example, based on the above-mentioned fuzzy and uncertain search terms, POIs that meet the real search needs of the search subject can be found, thereby meeting the specific and differentiated search needs of the search subject. In this embodiment, it is first necessary to determine the training strategy data corresponding to the POI search model to be trained, and the training strategy data matches the specific and differentiated real search needs in the specific search scenario targeted by the POI search model.
[0047] In this embodiment, when determining the training strategy data for the above-mentioned specific and differentiated search needs, it is first necessary to determine the degree of match between the search term data and the POI recommendation data based on the search term data input by the user in the historical POI search scenario, the POI recommendation data output to the user, and the user's selection results for the POI recommendation data, and determine the training strategy data corresponding to the POI search model to be trained based on the degree of match, that is, the training strategy data in this embodiment can be determined based on the user's actual POI search needs in a specific search scenario.
[0048] Based on this, this embodiment needs to determine the training strategy data corresponding to the POI search model to be trained in the following manner: first, obtain the user's actual search demand data for a specific search scenario; second, based on the actual search demand data, determine the training strategy data corresponding to the POI search model to be trained.
[0049] Specifically, this embodiment mainly determines the training strategy data corresponding to the POI search model to be trained for the following specific POI search scenarios:
[0050] For the POI search scenario (scenario one) in which the first search term includes a popular address name, the popular address name is inconsistent with the address name of the first POI targeted by the first search term, and the geographical location of the popular address name is within a predetermined distance range of the first POI, the corresponding real search demand data is the ability to search for the first POI based on the first search term; in this scenario, the training strategy data corresponding to the POI search model to be trained is determined, which specifically may refer to: performing text convergence model training based on the above-mentioned popular address names and the address name of the first POI, so that the trained POI search model can output the first POI based on the first search term; wherein the popular address name represents an address name for which the user's attention exceeds a predetermined attention threshold. For example, in the search scenario involving popular address names (such as shopping malls, schools, scenic spots, etc.) near the POI, users are often not clear about the full name of the POI they are searching for, but are more aware of the popular address names near the POI (for example, there is a Route XX supercharging station near Square A. When the search keyword is "Square A supercharging station", the user's actual search demand is to return to the Route XX supercharging station). Therefore, the model training strategy data for this scenario is to assign a higher text similarity between the address name of the POI and the popular address names around it, and form a training data set for model training based on this.
[0051] For the POI search scenario in which the second search term includes a geographical term (scenario two), the corresponding actual search demand data is: searching for POIs whose correlation with the above-mentioned geographical term exceeds a predetermined correlation threshold based on the second search term; the training strategy data corresponding to the determined POI search model to be trained may refer to: performing model training based on the correlation between the geographical term and the second POI corresponding to the second search term, so that the trained POI search model can output POIs whose correlation with the geographical term exceeds a predetermined correlation threshold based on the second search term.
[0052] For the POI search scenario (scenario three) where the third search term and its corresponding third POI are synonyms or homophones, the corresponding real search demand data is: searching for the third POI based on the third search term; correspondingly, the training strategy data corresponding to the determined POI search model to be trained may refer to: training the text convergence model based on the third search term and the third POI, so that the trained POI search model can output the third POI based on the third search term. For example, with the popularity of voice search, users are more inclined to propose search terms in a colloquial way. The model training strategy for homophones can better adapt to accent differences and ensure that the results obtained through voice search are more accurate; for another example, the user's input is diverse due to input method selection, personal habits, input errors, accidental touches, etc., and the pinyin or phrases that are homophonic with the search keyword are used. Considering the search strategy for homophone search scenarios helps to improve the intelligence and user-friendliness of the search engine and enable the model to understand the user's true intention from ambiguous or similar inputs. Training for synonym scenarios improves search accuracy and user experience. For another example, the search keywords entered by users may be ambiguous or have a wide semantic range. Different users may use different words to express the same concept. For example, users searching for "bus station" often also point to "bus passenger station". Therefore, by covering the synonym model training strategy, the model can adapt to the different word usage habits of users, improve the search coverage, and return richer and more comprehensive search results.
[0053] S102, obtaining search word samples and POI text samples corresponding to the training strategy data, and obtaining text similarity between the search word samples and the POI text samples.
[0054] After the training strategy data corresponding to the POI search model to be trained is determined in the above step, this step is used to obtain the search term samples and POI text samples corresponding to the above training strategy data, and obtain the text similarity between the search term samples and the POI text samples. The search term samples are the request texts in the training samples during model training, and the POI text samples are the target texts in the training samples during model training. The search term samples and POI text samples corresponding to the above training strategy data refer to: search term samples and POI text samples corresponding to the above specific search scenarios and actual search needs. Consistent with the process of determining the training strategy data in the above step S101, in the present embodiment, the search term samples and POI text samples corresponding to the training strategy data can be obtained from the preset historical search data corresponding to each training strategy data. Specifically, the degree of match between the search term data and the POI recommendation data is determined based on the search term data input by the user in the historical POI search scenario, the POI recommendation data output to the user, and the user's selection result for the POI recommendation data. If the degree of match between the search term data and the POI recommendation data exceeds a predetermined matching threshold, the search term data is used as a search term sample, and the POI recommendation data is used as a POI text sample.
[0055] For the POI search scenario in which the first search term in step S101 contains a popular address name, the popular address name is inconsistent with the address name of the first POI targeted by the first search term, and the geographical location of the popular address name is within a predetermined distance range of the first POI, the first search term containing the popular address name may be used as a search term sample, and the first POI may be used as a POI text sample;
[0056] For a POI search scenario where the second search term includes a geographical term, the second search term including the geographical term may be used as a search term sample, and the second POI may be used as a POI text sample;
[0057] For a POI search scenario in which the third search term and its corresponding third POI are synonyms or homophones, the third search term may be used as a search term sample, and the third POI may be used as a POI text sample.
[0058] In this embodiment, the text similarity between the search term sample and the POI text sample can be obtained in the following manner:
[0059] Method 1: The text similarity between the search word sample and the POI text sample is calculated based on a preset text similarity calculation method; for example, after collecting the name, category, description, user evaluation and other text data of the POI text information from the POI search scenario corresponding to the above training strategy data, the text is preprocessed by text cleaning, word segmentation, and stop word removal, and then features are extracted from the POI text information, including word embedding technologies such as Bag of Words, TF-IDF (TermFrequency-Inverse Document Frequency), Word2Vec, and BERT, and then text similarity is calculated based on the extracted features, using existing TF-IDF, multiple pre-trained embedded models, word vector text similarity algorithms (cosine similarity, Jaccard similarity, edit distance (Levenshtein Distance)), etc.
[0060] Method 2: Score the search term sample and the POI text sample according to the correlation between the search term and the POI text represented by the training strategy data, and obtain the text similarity between the search term sample and the POI text sample; for example, for the above scenario 1, the text similarity between the search term sample and the POI text sample is obtained according to the training strategy data, which may specifically refer to: in order to make the popular address name contained in the first search term in the POI search converge to the address name of the first POI, the first search term and the first POI are scored for text similarity, and a text similarity score representing that the text similarity between the first search term and the first POI is higher than a predetermined similarity threshold. For another example, for the above scenario 2, the text similarity between the search term sample and the POI text sample is obtained according to the training strategy data, which may specifically refer to: if the correlation between the geographical term contained in the second search term and the second POI exceeds the predetermined correlation, the second search term and the second POI are scored for text similarity, and a text similarity score representing that the text similarity between the second search term and the second POI is higher than the predetermined similarity threshold is obtained. For another example, for the above scenario three, the text similarity between the search term sample and the POI text sample is obtained according to the training strategy data, which may specifically mean: scoring the text similarity of the third search term and the third POI, and obtaining a text similarity score representing that the text similarity between the third search term and the third POI is higher than a predetermined similarity threshold.
[0061] Method 3: Calculate the initial similarity score between the search term sample and the POI text sample based on the preset text similarity calculation method, and adjust the initial similarity score according to the correlation between the search term and the POI text represented by the training strategy data to obtain the text similarity. For example, the initial similarity score with the lowest or highest score, the smallest variance or the largest variance can be removed.
[0062] S103, performing model training based on search word samples, POI text samples and text similarity to obtain a POI search model, where the POI search model is used to output a target POI text that matches the training strategy data according to the input target search word.
[0063] After obtaining the search word samples and POI text samples in the above steps, and obtaining the text similarity between the search word samples and the POI text samples, this step is used to use the data set consisting of the above search word samples, POI text samples and their text similarity to train the model, and use indicators such as accuracy, recall rate, F1 score, etc. to evaluate the model performance, and adjust the model parameters according to the evaluation results, and perform iterative training to improve the model performance. After the model training is completed, the trained POI search model can be deployed to the actual POI search system to provide real-time POI search services, and the POI search model can be continuously optimized according to user feedback and search logs to cope with subsequent continuously updated POI search needs and user needs.
[0064] The training method of the POI search model provided in the embodiment of the present application includes: determining the training strategy data corresponding to the POI search model to be trained; obtaining the search word samples and POI text samples corresponding to the training strategy data, and obtaining the text similarity between the search word samples and the POI text samples; performing model training based on the search word samples, POI text samples and text similarity to obtain the POI search model, which is used to output the target POI text matching the training strategy data according to the input target search word. The POI search model obtained by training using the present invention can be combined with the characteristics of the POI search scene to achieve accurate matching of the search word and the output POI text, so as to accurately identify the search intent of the POI search scene, thereby making the POI search results more accurate.
[0065] The above embodiment provides a training method for a POI search model. Correspondingly, another embodiment of the present application further provides a training device for a POI search model. Since the device embodiment is basically similar to the method embodiment, the description is relatively simple. For details of the relevant technical features, please refer to the corresponding description of the method embodiment provided above. The following description of the device embodiment is merely illustrative.
[0066] Please refer to Figure 2 To understand this embodiment, Figure 2 A unit block diagram of a training device for a POI search model provided in this embodiment, such as Figure 2 As shown, the training device of the POI search model provided in this embodiment includes:
[0067] A training strategy data determining unit 201 is used to determine the training strategy data corresponding to the POI search model to be trained;
[0068] A training sample obtaining unit 202, used to obtain search word samples and POI text samples corresponding to the training strategy data, and obtain text similarity between the search word samples and the POI text samples;
[0069] The model training unit 203 is used to perform model training based on search word samples, POI text samples and text similarity to obtain a POI search model, which is used to output a target POI text that matches the training strategy data based on the input target search word.
[0070] In some embodiments, the training strategy data corresponding to the POI search model to be trained is determined, including: for a POI search scenario in which a first search term contains a popular address name, the popular address name is inconsistent with the address name of the first POI targeted by the first search term, and the geographical location of the popular address name is within a predetermined distance range of the first POI, the training strategy data is determined as: text convergence model training is performed based on the popular address name and the address name of the first POI, so that the trained POI search model can output the first POI based on the first search term; wherein the popular address name represents an address name for which the user's attention exceeds a predetermined attention threshold.
[0071] In some implementations, obtaining the search term sample and the POI text sample corresponding to the training strategy data includes: taking the first search term containing the popular address name as the search term sample, and taking the first POI as the POI text sample;
[0072] The obtaining of the text similarity between the search term sample and the POI text sample according to the training strategy data includes: in order to make the popular address name contained in the first search term in the POI search converge to the address name of the first POI, scoring the text similarity of the first search term and the first POI, and obtaining a text similarity score representing that the text similarity between the first search term and the first POI is higher than a predetermined similarity threshold.
[0073] In some embodiments, the determination of training strategy data corresponding to the POI search model to be trained includes: for a POI search scenario in which a second search term contains a geographical term, performing model training based on the association between the geographical term and a second POI corresponding to the second search term, so that the trained POI search model can output a POI whose association with the geographical term exceeds a predetermined association threshold based on the second search term.
[0074] In some embodiments, the determination of training strategy data corresponding to the POI search model to be trained includes: for a POI search scenario in which a third search term and its corresponding third POI are synonyms or homophones, performing text convergence model training based on the third search term and the third POI, so that the trained POI search model can output the third POI based on the third search term.
[0075] In some embodiments, obtaining the text similarity between the search term sample and the POI text sample according to the training strategy data includes: calculating the text similarity score between the search term sample and the POI text sample based on a preset text similarity calculation method; and / or
[0076] According to the correlation between the search term and the POI text represented by the training strategy data, the text similarity between the search term sample and the POI text sample is scored or adjusted.
[0077] In some embodiments, determining the training strategy data corresponding to the POI search model to be trained includes: determining the degree of matching between the search term data and the POI recommendation data based on the search term data input by the user in the historical POI search scenario, the POI recommendation data output to the user, and the user's selection result for the POI recommendation data, and determining the training strategy data corresponding to the POI search model to be trained based on the matching degree.
[0078] The embodiment of the present invention further provides an electronic device, which can be provided with the above-mentioned POI search model training device in the form of a program to execute the POI search model training method provided by the embodiment of the present invention. Optionally, an optional hardware structure of the electronic device can be as follows Figure 3 As shown, it includes: at least one processor 301, at least one communication interface 302, at least one memory 303 and at least one communication bus 304;
[0079] Optionally, the communication interface 302 may be an interface of a communication module, such as an interface of a GSM module;
[0080] The processor 301 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention.
[0081] The memory 303 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.
[0082] The memory 303 stores a program, and the processor 301 calls the program stored in the memory 303 to execute the training method of the POI search model provided in the embodiment of the present invention.
[0083] In the above embodiments, a training method for a POI search model, a training device for a POI search model, and an electronic device are provided. In addition, another embodiment of the present application further provides a computer-readable storage medium for implementing the training method for the above POI search model. The computer-readable storage medium embodiment provided in this application is described relatively simply. For the relevant parts, please refer to the corresponding description of the above method embodiment. The embodiments described below are merely exemplary. The computer-readable storage medium provided in this embodiment stores computer instructions, which, when executed by a processor, implement the following steps:
[0084] Determine the training strategy data corresponding to the POI search model to be trained;
[0085] Obtaining a search word sample and a POI text sample corresponding to the training strategy data, and obtaining a text similarity between the search word sample and the POI text sample;
[0086] Model training is performed based on the search word samples, the POI text samples and the text similarity to obtain the POI search model, which is used to output a target POI text that matches the training strategy data according to an input target search word.
[0087] In some embodiments, the training strategy data corresponding to the POI search model to be trained is determined, including: for a POI search scenario in which a first search term contains a popular address name, the popular address name is inconsistent with the address name of the first POI targeted by the first search term, and the geographical location of the popular address name is within a predetermined distance range of the first POI, the training strategy data is determined as: text convergence model training is performed based on the popular address name and the address name of the first POI, so that the trained POI search model can output the first POI based on the first search term; wherein the popular address name represents an address name for which the user's attention exceeds a predetermined attention threshold.
[0088] In some implementations, obtaining the search term sample and the POI text sample corresponding to the training strategy data includes: taking the first search term containing the popular address name as the search term sample, and taking the first POI as the POI text sample;
[0089] The obtaining of the text similarity between the search term sample and the POI text sample according to the training strategy data includes: in order to make the popular address name contained in the first search term in the POI search converge to the address name of the first POI, scoring the text similarity of the first search term and the first POI, and obtaining a text similarity score representing that the text similarity between the first search term and the first POI is higher than a predetermined similarity threshold.
[0090] In some embodiments, the determination of training strategy data corresponding to the POI search model to be trained includes: for a POI search scenario in which a second search term contains a geographical term, performing model training based on the association between the geographical term and a second POI corresponding to the second search term, so that the trained POI search model can output a POI whose association with the geographical term exceeds a predetermined association threshold based on the second search term.
[0091] In some embodiments, the determination of training strategy data corresponding to the POI search model to be trained includes: for a POI search scenario in which a third search term and its corresponding third POI are synonyms or homophones, performing text convergence model training based on the third search term and the third POI, so that the trained POI search model can output the third POI based on the third search term.
[0092] In some embodiments, obtaining the text similarity between the search term sample and the POI text sample according to the training strategy data includes: calculating the text similarity score between the search term sample and the POI text sample based on a preset text similarity calculation method; and / or
[0093] According to the correlation between the search term and the POI text represented by the training strategy data, the text similarity between the search term sample and the POI text sample is scored or adjusted.
[0094] In some embodiments, determining the training strategy data corresponding to the POI search model to be trained includes: determining the degree of matching between the search term data and the POI recommendation data based on the search term data input by the user in the historical POI search scenario, the POI recommendation data output to the user, and the user's selection result for the POI recommendation data, and determining the training strategy data corresponding to the POI search model to be trained based on the matching degree.
[0095] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.
[0096] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.
[0097] 1. Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be used to store information by any method or technology. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include non-transitory media such as modulated data signals and carrier waves.
[0098] 2. Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0099] Although the present application is disclosed as above in the form of a preferred embodiment, it is not intended to limit the present application. Any technical personnel in this field may make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be based on the scope defined by the claims of the present application.
Claims
1. A training method for a POI search model, characterized in that: include: Determine the training strategy data corresponding to the POI search model to be trained; Obtaining a search term sample and a POI text sample corresponding to the training strategy data, and obtaining a text similarity between the search term sample and the POI text sample; Model training is performed based on the search word samples, the POI text samples and the text similarity to obtain the POI search model, which is used to output a target POI text that matches the training strategy data according to an input target search word.
2. The method according to claim 1, characterized in that The step of determining the training strategy data corresponding to the POI search model to be trained includes: Obtain users' real search demand data for specific search scenarios; Based on the actual search demand data, training strategy data corresponding to the POI search model to be trained is determined.
3. The method according to claim 2, characterized in that The specific search scenario includes: a POI search scenario in which a first search term includes a hot address name, the hot address name is inconsistent with the address name of a first POI targeted by the first search term, and the geographical location of the hot address name is within a predetermined distance range of the first POI; The actual search demand data includes: searching for the first POI based on the first search term; The determining of the training strategy data corresponding to the POI search model to be trained includes: performing text convergence model training based on the popular address name and the address name of the first POI, so that the trained POI search model can output the first POI based on the first search term; wherein the popular address name represents an address name for which the user's attention exceeds a predetermined attention threshold; The obtaining of the search term sample and the POI text sample corresponding to the training strategy data includes: taking the first search term containing the popular address name as the search term sample, and taking the first POI as the POI text sample.
4. The method according to claim 2, characterized in that: The specific search scenario includes: a POI search scenario in which the second search term includes a geographical term; The real search demand data includes: searching, based on the second search term, for POIs whose relevance to the geographical term exceeds a predetermined relevance threshold; The determining of the training strategy data corresponding to the POI search model to be trained includes: performing model training based on the association between the geographical term and the second POI corresponding to the second search term, so that the trained POI search model can output a POI whose association with the geographical term exceeds a predetermined association threshold based on the second search term; The obtaining of the search term sample and the POI text sample corresponding to the training strategy data includes: taking the second search term containing the geographical term as the search term sample, and taking the second POI as the POI text sample.
5. The method according to claim 2, characterized in that: The specific search scenario includes: a POI search scenario in which the third search term and its corresponding third POI are synonyms or homophones; The actual search demand data includes: searching for the third POI based on the third search term; The determining of the training strategy data corresponding to the POI search model to be trained includes: performing text convergence model training based on the third search term and the third POI, so that the trained POI search model can output the third POI based on the third search term; The obtaining of the search term sample and the POI text sample corresponding to the training strategy data includes: taking the third search term as the search term sample and taking the third POI as the POI text sample.
6. The method according to claim 1, characterized in that The obtaining the text similarity between the search term sample and the POI text sample according to the training strategy data comprises one of the following: Calculating the text similarity between the search term sample and the POI text sample based on a preset text similarity calculation method; Scoring the search term samples and the POI text samples according to the correlation between the search term and the POI text represented by the training strategy data to obtain the text similarity between the search term samples and the POI text samples; An initial similarity score between the search term sample and the POI text sample is calculated based on a preset text similarity calculation method, and the initial similarity score is adjusted according to the correlation between the search term and the POI text represented by the training strategy data to obtain the text similarity.
7. The method according to claim 1, characterized in that The determining of the training strategy data corresponding to the POI search model to be trained includes: determining the degree of matching between the search word data and the POI recommendation data according to the search word data input by the user in the historical POI search scene, the POI recommendation data output to the user, and the selection result of the user for the POI recommendation data, and determining the training strategy data corresponding to the POI search model to be trained based on the matching degree; The obtaining of the search term samples and POI text samples corresponding to the training strategy data includes: in response to a matching degree between the search term data and the POI recommendation data exceeding a predetermined matching threshold, taking the search term data as the search term sample and taking the POI recommendation data as the POI text sample.
8. A training device for a POI search model, characterized in that: include: A training strategy data determining unit, used to determine the training strategy data corresponding to the POI search model to be trained; A training sample obtaining unit, used to obtain a search word sample and a POI text sample corresponding to the training strategy data, and obtain a text similarity between the search word sample and the POI text sample; A model training unit is used to perform model training based on the search word sample, the POI text sample and the text similarity to obtain the POI search model, wherein the POI search model is used to output a target POI text matching the training strategy data according to an input target search word.
9. An electronic device, characterized in that: comprising a processor and a memory; wherein, The memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement the method according to any one of claims 1 to 7.
10. A computer-readable storage medium having one or more computer instructions stored thereon, characterized in that: The instructions are executed by a processor to implement the method according to any one of claims 1 to 7.