Search word completion suggestion method and device, equipment and medium
By calling the sequence labeling model and preset sorting model in real time in the address search scenario, search term completion suggestions are provided, and the results are displayed based on the similarity, the problem of inaccurate search term completion suggestions in the existing technology is solved, and the user experience and search term coverage are improved.
Patent Information
- Application Number
- CN202510203018.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-24
- Publication Date
- 2025-06-17
AI Technical Summary
The prior art is difficult to accurately provide search term completion suggestions in address search scenarios, resulting in extremely poor user experience. Especially in the search behavior of long-tail distribution, the prior art is difficult to cover all possible search terms.
The user input content is recalled by calling the sequence annotation model in real time, and the search term completion suggestions are output in combination with the preset sorting model, and the strategy is displayed based on the similarity between the search results of the address interest point and the completion suggestions.
It realizes more accurate search term completion suggestions, improves the user experience, effectively covers the search terms in the long-tail distribution, and improves the overall experience of user retrieval behavior.
Smart Images
Figure CN120163243A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of data processing, and particularly to a search term completion suggestion method, device, equipment and medium. Background Art
[0002] POI is the abbreviation of "Point of Interest", which can be translated as point of interest, and some are also called "Point of Information", that is, information point. POI point of interest data is the core data based on location services and is widely used in the scenario of electronic maps. For example, the destination selected by the user before navigation.
[0003] Taking the address search scenario as an example, when providing a search service, since the information input by the user in the search box is often incomplete, automatic completion is generally required, and then the completed information is displayed in the candidate area for the user to select, so as to find the required point of interest faster. The retrieval behavior of users shows a long-tail distribution. However, the existing technology solutions for completion methods have not specifically addressed this problem, bringing an extremely negative experience to users. Therefore, there is an urgent need for a method for search term completion suggestions that can accurately obtain search term completion suggestions based on the input content, improving the user experience. Summary of the Invention
[0004] The present application provides a search term completion suggestion method, device, equipment and medium, which can realize relatively accurate search term completion suggestions and improve the user experience.
[0005] In a first aspect, the present application provides a method for search term completion suggestions, the method comprising:
[0006] In response to a user input operation, real-time obtain the input content and the address point of interest search result corresponding to the input content, where the input content is a search term in an address search scenario;
[0007] Real-time call a sequence labeling model to recall the input content and obtain a recall result;
[0008] Input the recall result and the input content into a preset sorting model, and output a search term completion suggestion;
[0009] If the similarity between the address point of interest search result and the search term completion suggestion is greater than a preset similarity threshold, display the address point of interest search result;
[0010] If the similarity between the address point of interest search result and the search term completion suggestion is not greater than the preset similarity threshold, display the address point of interest search result and the search term completion suggestion.
[0011] Optionally, the real-time calling of the sequence annotation model to recall the input content to obtain a recall result includes:
[0012] The sequence labeling model is called in real time to perform component analysis on the input content to obtain a parsing result; the sequence labeling model is generated by training a Bert pre-training model based on a training data set after manual component labeling in an address search scenario;
[0013] Performing component judgment on the analysis result;
[0014] If the analysis result indicates that the input content does not contain any non-Chinese character components, the input content is recalled according to the analysis result to obtain the recall result;
[0015] If the analysis result indicates that the input content contains non-Chinese character components, the input content is recognized to obtain a recognition result, and recall is performed based on the input content and the recognition result to obtain the recall result.
[0016] Optionally, recalling the input content according to the parsing result to obtain the recall result includes:
[0017] When the parsing result indicates that the input content includes core words but does not include geographical component information, core word recall is performed on the input content to obtain the recall result.
[0018] Optionally, recalling the input content according to the parsing result to obtain the recall result includes:
[0019] When the analysis result indicates that the input content includes geographic component information and core words, the geographic component information in the input content is removed to obtain the input content after the removal;
[0020] Perform core word recall on the input content after the elimination to obtain the recall result.
[0021] Optionally, recalling the input content according to the parsing result to obtain the recall result includes:
[0022] When the parsing result indicates that the input content includes geographical component information but does not include core words, the geographical component is recalled for the input content to obtain the recall result.
[0023] Optionally, the recalling according to the input content and the recognition result to obtain the recall result includes:
[0024] If the recognition result indicates that the last character in the input content is a letter, perform letter recall on the last character in the input content to obtain the recall result.
[0025] Optionally, the recalling according to the input content and the recognition result to obtain the recall result includes:
[0026] If the recognition result indicates that the last character in the input content is a stroke, remove the stroke in the input content to obtain the input content after removal;
[0027] Perform recall on the input content after removal to obtain the recall result.
[0028] In a second aspect, the present application further provides a device for search term completion suggestions, and the device includes:
[0029] An acquisition unit, configured to respond to a user input operation, and acquire in real time the input content and the address point of interest search result corresponding to the input content, where the input content is a search term in an address search scenario;
[0030] A recall unit, configured to call a sequence labeling model in real time to perform recall on the input content to obtain a recall result;
[0031] An output unit, configured to input the recall result and the input content into a preset sorting model, and output search term completion suggestions;
[0032] A display unit, configured to display the address point of interest search result if the similarity between the address point of interest search result and the search term completion suggestions is greater than a preset similarity threshold;
[0033] The display unit is further configured to display the address point of interest search result and the search term completion suggestions if the similarity between the address point of interest search result and the search term completion suggestions is not greater than the preset similarity threshold.
[0034] Optionally, the recall unit includes:
[0035] A content parsing module, configured to call the sequence labeling model in real time to perform component parsing on the input content to obtain a parsing result; the sequence labeling model is trained and generated from a Bert pre-training model based on a training data set with manual component labeling in an address search scenario;
[0036] A component judgment module, configured to perform component judgment on the parsing result;
[0037] A first recall module, configured to recall the input content according to the analysis result to obtain the recall result if the analysis result indicates that the input content does not contain any non-Chinese character components;
[0038] The second recall module is used to recognize the input content to obtain a recognition result if the analysis result indicates that the input content contains non-Chinese character components, and to recall the input content based on the input content and the recognition result to obtain the recall result.
[0039] Optionally, the first recall module is specifically used to: when the parsing result indicates that the input content includes core words but does not include geographical component information, perform core word recall on the input content to obtain the recall result.
[0040] Optionally, the first recall module is specifically configured to:
[0041] When the analysis result indicates that the input content includes geographic component information and core words, the geographic component information in the input content is removed to obtain the input content after the removal;
[0042] Perform core word recall on the input content after the elimination to obtain the recall result.
[0043] Optionally, the first recall module is specifically configured to:
[0044] When the parsing result indicates that the input content includes geographical component information but does not include core words, the geographical component is recalled for the input content to obtain the recall result.
[0045] Optionally, the second recall module is specifically configured to:
[0046] If the recognition result indicates that the last character in the input content is a letter, a letter recall is performed on the last character in the input content to obtain the recall result.
[0047] Optionally, the second recall module is specifically configured to:
[0048] If the recognition result indicates that the last character in the input content is a stroke, then the strokes in the input content are removed to obtain the input content after the removal;
[0049] The input content after the elimination is recalled to obtain the recall result.
[0050] In a third aspect, the present application further provides an electronic device, the electronic device comprising a processor and a memory:
[0051] The memory is used to store computer programs;
[0052] The processor is configured to execute the method provided in the first aspect above according to the computer program.
[0053] In a fourth aspect, the present application further provides a computer-readable storage medium for storing a computer program for executing the method provided in the first aspect above.
[0054] It can be seen that the present application has the following beneficial effects:
[0055] The present application provides a search term completion suggestion method, apparatus, device, and medium. In this method, in response to a user input operation, the content that has been input and the address point of interest search results corresponding to the content that has been input are obtained in real time, and the content that has been input is a search term in an address search scenario; a sequence labeling model is called in real time to recall the content that has been input, and a recall result is obtained; the recall result and the content that has been input are input into a preset sorting model, and a search term completion suggestion is output; if the similarity between the address point of interest search result and the search term completion suggestion is greater than a preset similarity threshold, the address point of interest search result is displayed; if the similarity between the address point of interest search result and the search term completion suggestion is not greater than the preset similarity threshold, the address point of interest search result and the search term completion suggestion are displayed. In this way, through the method provided in the embodiments of the present application, a sequence labeling model can be called in real time to recall the content that has been input, a recall result is obtained, then the recall result and the content that has been input are input into a preset sorting model, a search term completion suggestion is output, the address point of interest search results corresponding to the content that has been input are obtained, and corresponding strategy displays are performed according to the similarity between the address point of interest search results and the search term completion suggestion. Because the sequence labeling model adopted in the embodiments of the present application can accurately recall according to the content that has been input, improving the coverage rate of the search term completion suggestion, it can more accurately perform search term completion suggestions for the content that the user has input, improving the user experience. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments recorded in the present application, and those of ordinary skill in the art can also obtain other drawings based on these drawings.
[0057] Figure 1 It is a flowchart of a search term completion suggestion method in an embodiment of the present application;
[0058] Figure 2 It is a flowchart of an embodiment of a search term completion suggestion method in an embodiment of the present application;
[0059] Figure 3 A schematic diagram of the structure of a search term completion suggestion device 300 provided in an embodiment of the present application;
[0060] Figure 4 A schematic diagram of the structure of an electronic device 400 provided in an embodiment of the present application. DETAILED DESCRIPTION
[0061] The multiple involved in the embodiments of the present application means greater than or equal to two. It should be noted that in the description of the embodiments of the present application, the words "first", "second", etc. are only used for the purpose of distinguishing the description, and cannot be understood as indicating or implying relative importance, nor can they be understood as indicating or implying an order.
[0062] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the embodiments of the present application are further described in detail below in conjunction with the accompanying drawings and specific implementation methods. It is understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. In addition, it should be noted that, for ease of description, only the parts related to the present application are shown in the accompanying drawings, not all structures.
[0063] Taking the address search scenario as an example, when providing search services, since the information entered by the user in the search box is often incomplete, it is generally necessary to automatically complete it, and then display the completed information in the candidate area for the user to choose. The user's search behavior shows a long-tail distribution, however, the completion method of the existing technical solution has not yet solved this problem in a targeted manner, which brings an extremely negative experience to users.
[0064] The applicant found through research that there are two main reasons for the long-tail distribution. On the one hand, there are popular large cities, such as Beijing, Shanghai, Guangzhou and Shenzhen. Compared with some small cities, these cities have high overall POI search popularity and rich data. On the other hand, in the same city, there are also popular POIs, which have high search popularity, while unpopular POIs have low search popularity.
[0065] The above situation will cause the search for points of interest (POI) to present a long-tail distribution. Most of the location instance cases mined using existing technologies (such as log mining methods, template generation methods, etc.) are high-frequency location instance cases, and there is a high probability that the long-tail location instance cases will not be covered.
[0066] To facilitate the understanding of the long-tail distribution concept, an example is given below for illustration. Many of the mined instance cases are retrieval instance cases in Shanghai. When constructing the input "Huai", since the proportion of "Huaihai Middle Road" mined from history is high when inputting "Huai", it is completed to "Huaihai". However, the real intention of the user is "Huaiyang Beef Soup", and the proportion of such instance cases is relatively small, making it impossible to correctly complete the input according to the user's input, resulting in an extremely negative experience.
[0067] Based on this, the embodiments of the present application provide a search term completion suggestion method, device, equipment, and medium. In this method, in response to a user input operation, the content that has been input and the address point of interest search results corresponding to the content that has been input are obtained in real time. The content that has been input is a search term in an address search scenario; a sequence labeling model is called in real time to recall the content that has been input, and a recall result is obtained; the recall result and the content that has been input are input into a preset sorting model, and a search term completion suggestion is output; if the similarity between the address point of interest search result and the search term completion suggestion is greater than a preset similarity threshold, the address point of interest search result is displayed; if the similarity between the address point of interest search result and the search term completion suggestion is not greater than the preset similarity threshold, the address point of interest search result and the search term completion suggestion are displayed.
[0068] In this way, through the method provided by the embodiments of the present application, a sequence labeling model can be called in real time to recall the content that has been input, and a recall result is obtained. Then, the recall result and the content that has been input are input into a preset sorting model, and a search term completion suggestion is output. The address POI search result corresponding to the content that has been input is obtained, and corresponding strategies are displayed according to the similarity between the address POI search result and the search term completion suggestion. Since the sequence labeling model adopted by the embodiments of the present application can accurately recall according to the content that has been input, improving the coverage rate of the search term completion suggestion, it can more accurately provide search term completion suggestions for the content that the user has input, improving the user experience.
[0069] To facilitate the understanding of the specific implementation of the search term completion suggestion method provided by the embodiments of the present application, the following will be described with reference to the accompanying drawings.
[0070] It should be noted that the main body implementing the search term completion suggestion method may be the search term completion suggestion device provided by the embodiments of the present application, and the search term completion suggestion device may be carried on an electronic device or a functional module of an electronic device. The electronic device in the embodiments of the present application can be any device capable of implementing the search term completion suggestion method in the embodiments of the present application, such as an Internet of Things (IoT) device.
[0071] See Figure 1 , which provides a flowchart of a search term completion suggestion method. In the embodiments of the present application, for example, the following steps may be included:
[0072] S1, in response to a user input operation, obtain the input content and the corresponding address point of interest search results in real time.
[0073] It should be noted that the input content can be a search term in an address search scenario. In a specific implementation process, an address search scenario corresponds to a database that stores the corresponding relationship between historical input content and address point of interest search results. Then, the corresponding address point of interest search results can be obtained according to the input content.
[0074] S2, call the sequence labeling model in real time to recall the input content and obtain the recall results.
[0075] It should be noted that the sequence labeling model is trained by using a training data set labeled with artificial components in an address search scenario to train the Bert (Bidirectional Encoder Representations from Transformers) pre-trained model. Among them, Bert is a pre-trained language model based on the Transformer architecture.
[0076] In a specific implementation process, the training data set stores user historical behavior data, which can include user historical input content and user historical click data. First, use the user historical click data for artificial component labeling. The label types include 29 labels such as province, city, road, area, business district - entity word, area, business district - category word, community - entity word, community - category word, entity word, category word, etc. Then, use the training data set after artificial component labeling to train the Bert pre-trained model. Among them, the core word is the entity word among them.
[0077] In a possible implementation manner, S2 provided in the embodiments of the present application may include:
[0078] S21, call the sequence labeling model in real time to perform component parsing on the input content and obtain the parsing results.
[0079] Because the sequence labeling model is trained by using a training data set labeled with artificial components in an address search scenario to train the Bert pre-trained model, the sequence labeling model can identify the core word and geographical components. Among them, the core word and geographical components obtained by parsing historical behavior data during the training of the sequence labeling model are stored in the recall library. When the sequence labeling model is called later, the recall library can be directly called to complement the core information and geographical component information in the user input content. Moreover, in order to achieve fast recall, the recall library can include a core word recall library and a geographical component recall library.
[0080] S22, perform component judgment on the parsing results.
[0081] S23, if the analysis result indicates that the input content does not contain any non-Chinese character components, the input content is recalled according to the analysis result to obtain a recall result.
[0082] In the embodiment of the present application, when the analysis result indicates that the input content does not contain non-Chinese character components, different recall strategies are adopted according to the specific analysis result.
[0083] In a possible implementation, S23 may include: when the parsing result indicates that the input content includes core words but does not include geographic component information, performing core word recall on the input content to obtain a recall result.
[0084] In another possible implementation, S23 may include: when the parsing result indicates that the input content includes geographic component information and core words, removing the geographic component information from the input content to obtain the input content after the removal; and performing core word recall on the input content after the removal to obtain a recall result.
[0085] In another possible implementation, S23 may include: when the parsing result indicates that the input content includes geographic component information but does not include core words, recalling the geographic component of the input content to obtain a recall result.
[0086] In the specific implementation process, the geographic component information can be, for example, administrative divisions. When the input content only includes administrative divisions and the administrative divisions are incomplete, the administrative division level of the input part is used to perform geographic component prefix matching or recall. Then, when the input content is complete (that is, it includes relatively complete geographic component information and core words), the core word recall is performed on the part of the input content excluding the administrative divisions. When the input content only contains the core word, that is, the user directly enters the core word, then the core word recall is performed.
[0087] S24, if the analysis result indicates that the input content contains non-Chinese character components, the input content is recognized to obtain a recognition result, and recall is performed based on the input content and the recognition result to obtain a recall result.
[0088] It should be noted that, in the embodiment of the present application, the analysis result is required to indicate that the input content contains non-Chinese character components and the last character in the input content is a non-Chinese character component.
[0089] That is to say, when the parsed result indicates that the input content contains non-Chinese characters, it is necessary to identify the input content to obtain the identification result, and then adopt different recall strategies according to the identification result. Since letters and strokes in the user input content will be identified as other components during the training process of the sequence labeling model, when using this sequence labeling model, not only can the input content be recalled for geographical components and core words, but also letter recall (such as letter prefix recall or letter abbreviation recall) can be performed. If the last component is identified as other, then it is determined whether the other component is a letter through regular matching and whether it is a stroke through the Chinese stroke dictionary.
[0090] It should be noted that when the parsed result indicates that the input content contains non-Chinese characters and the last character in the input content is a non-Chinese character, the input content is identified to obtain the identification result. Since directly using letter recall of pinyin letters will cause over-recall problems because there are a large number of results with the same pinyin letter prefix, and as the user continuously inputs, the combinations can be Chinese, Chinese + pinyin letters, or Chinese + Chinese strokes. Therefore, the components of the parsed result should be judged first, and then corresponding recalls should be performed for different components.
[0091] In a possible implementation manner, if the identification result indicates that the last character in the input content is a letter, letter recall is performed on the last character in the input content to obtain the recall result.
[0092] In another possible implementation manner, if the identification result indicates that the last character in the input content is a stroke, the stroke in the input content is removed to obtain the input content after removal; recall is performed on the input content after removal to obtain the recall result.
[0093] In the embodiments of the present application, special processing is performed on non-Chinese characters, such as processing forms like incomplete pinyin and strokes, so as to more comprehensively and accurately perform search term completion suggestions.
[0094] It should be noted that the embodiments of the present application can call the sequence labeling model to parse the components of the input content, and specify corresponding recall strategies for different components, improving the coverage rate of the recalled search term completion suggestions. Therefore, it has a certain ability to solve the instances in the long tail.
[0095] It should be noted that the prior art only constructs a recall set by using the user's historical information. The recall strategy is simple, and the granularity of the recall set is not consistent. The results recalled by the prior art may be a highly complete point of interest (POI), a rather incomplete query that has been historically clicked, or may also be an irrelevant but partially similar POI, with great uncertainty. The recall results are sometimes good and sometimes bad. In contrast, the embodiments of the present application will simultaneously obtain the recall results of the input content and the POI search results, and call a sequence labeling model to analyze and recall the input content. Since the component diversity and form diversity of the input content are considered when constructing the sequence labeling model, the input content is disassembled in terms of granularity and form, taking into account the different granularities and forms of the input content by the user. Then, a more accurate recall result can be obtained through the sequence labeling model, thereby improving the accuracy of the obtained search term completion suggestions.
[0096] S3. Input the recall result and the input content into a preset sorting model, and output search term completion suggestions.
[0097] It should be noted that the preset sorting model is obtained by training a lightweight gradient boosting machine (LightGBM) model with historical input content as the input and historical click data as the output. The preset sorting model outputs a preset number of search term completion suggestions sorted in descending order. Among them, LightGBM is based on the gradient boosting framework and optimizes the objective function by iteratively constructing decision trees. In the sorting task, the objective function is usually a loss function related to sorting. The LightGBM model efficiently processes complex functions and continuously updates the model parameters through the gradient descent method, thereby improving the sorting performance. That is, a preset sorting model with good sorting performance can be obtained through the method for obtaining the preset sorting model provided by the embodiments of the present application.
[0098] Among them, the preset number can be set by the staff according to the actual situation. In the embodiments of the present application, in order to improve the user experience, the preset number can be set to no more than 2, that is, at most only 2 search term completion suggestions are returned to the user. If too many are displayed, then the difference between completion and retrieval is not significant, and the user still needs to distinguish and select by themselves, reducing the user experience. The embodiments of the present application sort the search term completion suggestions in descending order based on the score results, and select the two with the highest score results as the search term completion suggestions, so that these 2 search term completion suggestions contain the POIs of the user's true intention, enabling quick selection and click, and improving the user experience.
[0099] In the specific implementation process, the preset sorting model constructs relevant features based on the length of the user's historical input content, the user's historical search behavior, and the user's historical click behavior, and is obtained by training a lightGBM model with the user's historical click behavior as the target output, and can output a preset number of search term completion suggestions sorted in descending order.
[0100] It should be noted that there are many results completed by using the existing completion methods. It may be a complete point of interest (POI), a similar query that the user has input in the past, or a POI with the same few characters but completely different from the user's true intention. It is difficult to determine an appropriate suggestion granularity in the prior art. Here, the suggestion granularity refers to the length after completion, and can be understood as whether it is a complete POI after completion or an incomplete POI. In the embodiments of the present application, the preset model constructs relevant features based on the length of the user's historical input content, the user's historical search behavior, and the user's historical click behavior, and is obtained by training a lightGBM model with the user's historical click behavior as the target output. Because it fits the relevant features constructed from the length of the user's historical input content, the user's historical search behavior, and the user's historical click behavior, using the preset sorting model to sort the recall results can determine an appropriate suggestion granularity and improve the accuracy of the search term completion suggestions.
[0101] S4. If the similarity between the address POI search result and the search term completion suggestion is greater than the preset similarity threshold, display the address POI search result.
[0102] S5. If the similarity between the address POI search result and the search term completion suggestion is not greater than the preset similarity threshold, display the address POI search result and the search term completion suggestion.
[0103] It should be noted that in steps S4 - S5, when the similarity between the address POI search result and the search term completion suggestion is relatively high, it is very likely that the address POI search result is what the user needs, and there is no need to display the search term completion suggestion. Only displaying the address POI search result can improve the user experience. Of course, if the similarity between the address POI search result and the search term completion suggestion is not high, then display both the address POI search result and the search term completion suggestion so that the user can make a choice based on the search term completion suggestion, quickly complete the completion, and obtain the POI of the user's true intention, improving the user experience.
[0104] Among them, the similarity can be the search term completion suggestion in the second sorting position among the address POI search results and the output search term completion suggestions. Of course, this is only an example here, and the embodiments of the present application are not limited thereto.
[0105] Thus, through the method provided by the embodiments of the present application, the sequence annotation model can be called in real time to recall the input content, obtain the recall result, and then input the recall result and the input content into the preset sorting model to output the search term completion suggestion, obtain the address point of interest search result corresponding to the input content, and display the corresponding strategy according to the similarity between the address point of interest search result and the search term completion suggestion. Since the sequence annotation model adopted by the embodiments of the present application is generated by training the Bert pre-trained model with the training data set labeled with artificial components in the address search scenario, it can accurately recall according to the input content, has a certain ability to solve the instance cases in the long tail, improves the coverage rate of the search term completion suggestion, can more accurately provide the search term completion suggestion for the content input by the user, and improves the user experience.
[0106] To make the method provided by the embodiments of the present application clearer and easier to understand, the following will describe a specific example of the method in combination with Figure 1 the scenario.
[0107] As Figure 2 shown, the embodiments of the present application may include, for example:
[0108] A Obtain the user query.
[0109] Among them, ideally, the user query may include the core word and geographical component information.
[0110] B Analyze the components in the user query and perform a point of interest (POI) retrieval on the user query.
[0111] C Perform geographical component recall according to the geographical component information in the user query, and perform core word recall according to the core word in the user query content.
[0112] Among them, when the user query consists of geographical components and core words, different recall strategies are adopted according to different input stages.
[0113] In the initial stage of input (that is, when only geographical components exist and the geographical components are relatively incomplete), perform geographical component prefix matching or recall according to the administrative division level where the input part is located. After the geographical components are relatively complete, perform core word recall; when the user directly inputs the core word, perform core word recall; when it is recognized that there are letters in the user's input query content, perform recall such as pinyin prefix and abbreviation; when it is recognized that there are possible strokes in the user's input query content, remove the relevant characters and then perform recall.
[0114] D Obtain the top 2 search term completion suggestions according to the geographical component recall, core word recall, and user query.
[0115] Determine whether to display the search term completion suggestion with reference to the POI retrieval result.
[0116] In the embodiments of the present application, after the user inputs the query content, the point of interest (POI) retrieval and the search term completion suggestion are performed simultaneously. During the process of generating the search term completion suggestion, first, the geographical component and the core word of the user query are parsed. Secondly, according to the parsing result, recall is performed in the geographical component recall library and the core word recall library. Then, the recall results are sorted in descending order of the scoring results, and the top 2 are taken as the search term completion suggestions. Determine whether to display the search term completion suggestion with reference to the POI retrieval result.
[0117] See Figure 3 , the embodiments of the present application further provide a device 300 for search term completion suggestion. The device 300 includes:
[0118] An acquisition unit 301, configured to, in response to a user input operation, acquire the input content in real time and the address point of interest search result corresponding to the input content, where the input content is a search term in an address search scenario;
[0119] A recall unit 302, configured to recall the input content in real time by invoking a sequence annotation model to obtain a recall result;
[0120] An output unit 303, configured to input the recall result and the input content into a preset sorting model and output a search term completion suggestion;
[0121] A display unit 304, configured to display the address point of interest search result if the similarity between the address point of interest search result and the search term completion suggestion is greater than a preset similarity threshold;
[0122] The display unit 304 is further configured to display the address point of interest search result and the search term completion suggestion if the similarity between the address point of interest search result and the search term completion suggestion is not greater than the preset similarity threshold.
[0123] Optionally, the recall unit 302 includes:
[0124] A content parsing module, configured to parse the components of the input content in real time by invoking the sequence annotation model to obtain a parsing result; the sequence annotation model is generated by training a Bert pre-trained model based on a training data set with manual component annotation in an address search scenario;
[0125] A component judgment module, configured to judge the components of the parsing result;
[0126] A first recall module, configured to recall the input content according to the analysis result to obtain the recall result if the analysis result indicates that the input content does not contain any non-Chinese character components;
[0127] The second recall module is used to recognize the input content to obtain a recognition result if the analysis result indicates that the input content contains non-Chinese character components, and to recall the input content based on the input content and the recognition result to obtain the recall result.
[0128] Optionally, the first recall module is specifically used to: when the parsing result indicates that the input content includes core words but does not include geographical component information, perform core word recall on the input content to obtain the recall result.
[0129] Optionally, the first recall module is specifically configured to:
[0130] When the analysis result indicates that the input content includes geographic component information and core words, the geographic component information in the input content is removed to obtain the input content after the removal;
[0131] Perform core word recall on the input content after the elimination to obtain the recall result.
[0132] Optionally, the first recall module is specifically configured to:
[0133] When the parsing result indicates that the input content includes geographical component information but does not include core words, the geographical component is recalled for the input content to obtain the recall result.
[0134] Optionally, the second recall module is specifically configured to:
[0135] If the recognition result indicates that the last character in the input content is a letter, a letter recall is performed on the last character in the input content to obtain the recall result.
[0136] Optionally, the second recall module is specifically configured to:
[0137] If the recognition result indicates that the last character in the input content is a stroke, then the strokes in the input content are removed to obtain the input content after the removal;
[0138] The input content after the elimination is recalled to obtain the recall result.
[0139] In addition, the present application embodiment also provides an electronic device 400, such as Figure 4 As shown, the electronic device 400 includes a processor 401 and a memory 402:
[0140] The memory 402 is used to store computer programs;
[0141] The processor 401 is used to execute according to the computer program Figure 1 or Figure 2 the provided method.
[0142] In addition, an embodiment of the present application also provides a computer-readable storage medium, which is used to store a computer program, and the computer program is used to execute the method provided by the embodiment of the present application.
[0143] From the description of the above embodiments, those skilled in the art can clearly understand that all or part of the steps in the above embodiment methods can be implemented by means of software plus a general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, and the computer software product can be stored in a storage medium, such as a read-only memory (ROM) / RAM, magnetic disk, optical disc, etc., including several instructions to enable a computer device (which can be a personal computer, a server, or a network communication device such as a router) to execute the methods described in various embodiments or some parts of the embodiments of the present application.
[0144] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The device embodiments described above are only illustrative. The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the objectives of the embodiment solution. Those of ordinary skill in the art can understand and implement it without creative efforts.
[0145] The above is only the preferred embodiment of the present application and is not used to limit the protection scope of the present application. It should be noted that for those of ordinary skill in the art in the technical field of the present application, several improvements and refinements can be made without departing from the premise of the present application, and these improvements and refinements should also be regarded as the protection scope of the present application.
Claims
1. A method for providing search term completion suggestions, characterized in that: The method comprises: In response to a user input operation, obtaining input content and address point of interest search results corresponding to the input content in real time, wherein the input content is a search term in an address search scenario; Invoke the sequence labeling model in real time to recall the input content and obtain a recall result; Inputting the recall result and the input content into a preset sorting model, and outputting search term completion suggestions; If the similarity between the address point of interest search result and the search term completion suggestion is greater than a preset similarity threshold, displaying the address point of interest search result; If the similarity between the address point of interest search result and the search term completion suggestion is not greater than the preset similarity threshold, the address point of interest search result and the search term completion suggestion are displayed.
2. The method according to claim 1, characterized in that The real-time calling of the sequence annotation model to recall the input content to obtain a recall result includes: The sequence labeling model is called in real time to perform component analysis on the input content to obtain a parsing result; the sequence labeling model is generated by training a Bert pre-training model based on a training data set after manual component labeling in an address search scenario; Performing component judgment on the analysis result; If the analysis result indicates that the input content does not contain any non-Chinese character components, the input content is recalled according to the analysis result to obtain the recall result; If the analysis result indicates that the input content contains non-Chinese character components, the input content is recognized to obtain a recognition result, and recall is performed based on the input content and the recognition result to obtain the recall result.
3. The method according to claim 2, characterized in that The step of recalling the input content according to the parsing result and obtaining the recall result includes: When the parsing result indicates that the input content includes core words but does not include geographical component information, core word recall is performed on the input content to obtain the recall result.
4. The method according to claim 2, characterized in that: The step of recalling the input content according to the parsing result and obtaining the recall result includes: When the analysis result indicates that the input content includes geographic component information and core words, the geographic component information in the input content is removed to obtain the input content after the removal; Perform core word recall on the input content after the elimination to obtain the recall result.
5. The method according to claim 2, characterized in that: The step of recalling the input content according to the parsing result and obtaining the recall result includes: When the parsing result indicates that the input content includes geographical component information but does not include core words, the geographical component is recalled for the input content to obtain the recall result.
6. The method according to claim 2, characterized in that The recalling according to the input content and the recognition result to obtain the recall result includes: If the recognition result indicates that the last character in the input content is a letter, a letter recall is performed on the last character in the input content to obtain the recall result.
7. The method according to claim 2, characterized in that The recalling according to the input content and the recognition result to obtain the recall result includes: If the recognition result indicates that the last character in the input content is a stroke, then the strokes in the input content are removed to obtain the input content after the removal; The input content after the elimination is recalled to obtain the recall result.
8. A device for providing search term completion suggestions, characterized in that: The device comprises: An acquisition unit, configured to acquire input content and address point of interest search results corresponding to the input content in real time in response to a user input operation, wherein the input content is a search term in an address search scenario; A recall unit, used to call the sequence labeling model in real time to recall the input content and obtain a recall result; An output unit, used to input the recall result and the input content into a preset sorting model, and output search term completion suggestions; A display unit, configured to display the address point of interest search result if the similarity between the address point of interest search result and the search term completion suggestion is greater than a preset similarity threshold; The display unit is further configured to display the address point of interest search result and the search term completion suggestion if the similarity between the address point of interest search result and the search term completion suggestion is not greater than the preset similarity threshold.
9. An electronic device, characterized in that: The electronic device comprises a processor and a memory: The memory is used to store computer programs; The processor is configured to execute the method according to any one of claims 1 to 7 according to the computer program.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store a computer program, and the computer program is used to execute the method according to any one of claims 1 to 7.