Information search method, device, server and storage medium

Abstract

The present application discloses an information search method, device, server and storage medium, belonging to the field of Internet technology. The method includes: determining multiple entity word fragments corresponding to a search statement, wherein at least one entity word fragment among the multiple entity word fragments includes multiple entity words, and the entity words are entity words in the search statement; for each of the entity word fragments, determining the target scene label associated with the entity word fragment, obtaining multiple target scene labels, each of the target scene labels is linked to a search result; determining a first correlation parameter between each of the entity word fragments and the target scene label associated with the entity word fragment; based on the first correlation parameter corresponding to each of the target scene labels, determining a target search result from multiple search results, and sending the target search result to the terminal. The above scheme improves the accuracy of information search.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of Internet technology, and in particular to an information search method, device, server and storage medium. Background Art

[0002] Currently, users search for life services online through terminals, and then perform business processing based on the searched information, such as booking tickets, ordering meals, or booking cars, etc. Among them, the user enters a search statement through the terminal, the terminal sends the search statement to the server, the server processes the search statement, obtains search results, and sends the search results to the terminal, so that the user can view the search results through the terminal and perform subsequent business processing based on the search results.

[0003] In the related art, during the information search process, the server performs entity recognition on the search statement input by the user to obtain entity words, then expands the entity words, and determines the search results to be sent to the terminal based on the expanded entity words.

[0004] When there are many entity words recognized, the search results determined according to the above scheme are relatively scattered, that is, the association with the search statement is not close enough, so the accuracy of the determined search results is not high, that is, the accuracy of the information search is not high. Summary of the invention

[0005] The embodiments of the present application provide an information search method, device, server and storage medium, which can improve the accuracy of information search. The technical solution is as follows:

[0006] On the one hand, an embodiment of the present application provides an information search method, the method comprising:

[0007] Determine multiple entity word fragments corresponding to the search sentence, wherein at least one entity word fragment among the multiple entity word fragments includes multiple entity words, and the entity words are entity words in the search sentence;

[0008] For each entity word fragment, determining a target scene label associated with the entity word fragment, and obtaining a plurality of target scene labels, each of which is linked to a search result;

[0009] Determining a first correlation parameter between each of the entity word fragments and a target scene label associated with the entity word fragment;

[0010] Based on the first correlation parameter corresponding to each of the target scene tags, a target search result is determined from multiple search results, and the target search result is sent to the terminal.

[0011] In some embodiments, determining a plurality of entity word segments corresponding to the search statement includes:

[0012] Performing entity recognition on the search sentence to obtain multiple entity words;

[0013] The multiple entity words are combined according to the position of each entity word in the search sentence to obtain the multiple entity word fragments.

[0014] In some embodiments, determining a first correlation parameter between each of the entity word segments and a target scene label associated with the entity word segment includes:

[0015] For each of the entity word fragments, feature extraction is performed on the entity word fragment to obtain a first feature, feature extraction is performed on the target scene label to obtain a second feature, and a second correlation parameter between the first feature and the second feature is determined;

[0016] Based on the historical feedback parameter of the target scene label, the second correlation parameter is adjusted to obtain the first correlation parameter, and the historical feedback parameter is used to represent the feedback status of the target scene label in the historical search process.

[0017] In some embodiments, adjusting the second correlation parameter based on the historical feedback parameter of the target scene tag to obtain the first correlation parameter includes:

[0018] determining a first increment of the second correlation parameter based on a first difference between the historical feedback parameter and a feedback threshold, wherein the first difference is positively correlated with the first increment;

[0019] The sum of the first increment and the second correlation parameter is taken as the first correlation parameter.

[0020] In some embodiments, the method further comprises:

[0021] Based on a search notification message reported by any terminal, a historical feedback parameter of a scene tag linked to a historical search result is determined, wherein the search notification message is used to indicate feedback status of the terminal on the historical search result.

[0022] In some embodiments, determining a first correlation parameter between each of the entity word segments and a target scene label associated with the entity word segment includes:

[0023] For each of the entity word fragments, feature extraction is performed on the entity word fragment to obtain a first feature, feature extraction is performed on the target scene label to obtain a second feature, and a second correlation parameter between the first feature and the second feature is determined;

[0024] Based on the priority corresponding to the target scene tag, the second correlation parameter is adjusted to obtain the first correlation parameter, where the priority is the priority of the target retrieval method, and the target retrieval method is the retrieval method used to retrieve the target scene tag.

[0025] In some embodiments, adjusting the second correlation parameter based on the priority corresponding to the target scene tag to obtain the first correlation parameter includes:

[0026] determining a second increment of the second relevance parameter based on a second difference between the priority and the priority threshold, the second difference being positively correlated with the second increment;

[0027] The sum of the second increment and the second correlation parameter is used as the first correlation parameter.

[0028] In some embodiments, the method further comprises:

[0029] Acquire multiple scene tags, where the scene tags are used to represent scene requirements to be searched;

[0030] For each of the scene tags, extract a core word from the scene tag, and determine entity mention information and a reference search statement of the scene tag based on the core word, wherein the entity mention information includes at least one entity mention related to the core word, and the reference search statement is a historical search statement related to the core word;

[0031] Based on the scene label, the core word, the entity mention information and the reference search sentence, a scene label knowledge base is constructed, wherein the scene label knowledge base is used to store the corresponding relationships between the scene label and the core word, the entity mention information and the reference search sentence respectively;

[0032] For each entity word segment, determining a target scene label associated with the entity word segment to obtain multiple target scene labels includes:

[0033] For each of the entity word fragments, a target scene label associated with the entity word fragment is determined from the scene label knowledge base to obtain the multiple target scene labels.

[0034] In some embodiments, determining a target search result from a plurality of search results based on the first relevance parameter corresponding to each of the target scene tags includes:

[0035] Based on the first correlation parameter, the multiple target scene tags are sorted to obtain a plurality of sorted target scene tags; from the multiple search results, the search results linked to the multiple target scene tags sorted before the target number are used as the target search results; or

[0036] The multiple search results are used as the target search results.

[0037] In some embodiments, sending the target search result to the terminal includes:

[0038] The target search results are sent to the terminal in sequence according to the sorted order of the multiple target scene tags.

[0039] In some embodiments, the plurality of target scene tags include a plurality of levels, each level includes at least one target scene tag, and the target scene tag in the next level is a sub-scene tag of the target scene tag in the previous level;

[0040] Before determining a target search result from a plurality of search results based on the first correlation parameter corresponding to each of the target scene tags, the method further includes:

[0041] From the target scene labels that have a hierarchical relationship, the target scene labels belonging to the previous level are deleted to obtain multiple updated target scene labels.

[0042] In some embodiments, determining a target search result from a plurality of search results based on the first relevance parameter corresponding to each of the target scene tags includes:

[0043] For each of the target scene tags, determining a third correlation parameter between the target scene tag and the search statement;

[0044] The target search result is determined from the multiple search results based on the first relevance parameter and the third relevance parameter corresponding to each of the target scene tags.

[0045] On the other hand, an embodiment of the present application provides an information search device, the device comprising:

[0046] A fragment determination module, used to determine a plurality of entity word fragments corresponding to a search sentence, wherein at least one entity word fragment among the plurality of entity word fragments includes a plurality of entity words, and the entity words are entity words in the search sentence;

[0047] A tag determination module, used for determining, for each entity word segment, a target scene tag associated with the entity word segment, to obtain a plurality of target scene tags, each of which is linked to a search result;

[0048] A first parameter determination module, used to determine a first correlation parameter between each of the entity word fragments and a target scene label associated with the entity word fragment;

[0049] The result determination module is used to determine a target search result from multiple search results based on a first correlation parameter corresponding to each target scene tag, and send the target search result to a terminal.

[0050] In some embodiments, the fragment determination module is used to perform entity recognition on the search statement to obtain multiple entity words; and combine the multiple entity words according to the position of each entity word in the search statement to obtain the multiple entity word fragments.

[0051] In some embodiments, the first parameter determination module includes:

[0052] A feature extraction unit is used to perform feature extraction on each entity word segment to obtain a first feature, perform feature extraction on the target scene label to obtain a second feature, and determine a second correlation parameter between the first feature and the second feature;

[0053] A parameter adjustment unit is used to adjust the second correlation parameter based on the historical feedback parameter of the target scene label to obtain the first correlation parameter, wherein the historical feedback parameter is used to represent the feedback status of the target scene label during the historical search process.

[0054] In some embodiments, the parameter adjustment unit is used to determine a first increased value of the second correlation parameter based on a first difference between the historical feedback parameter and a feedback threshold, wherein the first difference is positively correlated with the first increased value; and take the sum of the first increased value and the second correlation parameter as the first correlation parameter.

[0055] In some embodiments, the apparatus further comprises:

[0056] The second parameter determination module is used to determine the historical feedback parameter of the scene tag linked to the historical search results based on the search notification message reported by any terminal, and the search notification message is used to indicate the feedback of the terminal to the historical search results.

[0057] In some embodiments, the first parameter determination module includes:

[0058] A feature extraction unit is used to perform feature extraction on each entity word segment to obtain a first feature, perform feature extraction on the target scene label to obtain a second feature, and determine a second correlation parameter between the first feature and the second feature;

[0059] The parameter adjustment unit is also used to adjust the second correlation parameter based on the priority corresponding to the target scene tag to obtain the first correlation parameter, wherein the priority is the priority of the target retrieval method, and the target retrieval method is the retrieval method used to retrieve the target scene tag.

[0060] In some embodiments, the parameter adjustment unit is further used to determine a second increased value of the second correlation parameter based on a second difference between the priority and the priority threshold, the second difference being positively correlated with the second increased value; and take the sum of the second increased value and the second correlation parameter as the first correlation parameter.

[0061] In some embodiments, the apparatus further comprises:

[0062] A knowledge base construction module is used to obtain multiple scene tags, where the scene tags are used to represent the scene requirements to be searched; for each of the scene tags, a core word is extracted from the scene tag, and based on the core word, entity mention information and a reference search statement of the scene tag are determined, where the entity mention information includes at least one entity mention related to the core word, and the reference search statement is a historical search statement related to the core word; based on the scene tag, the core word, the entity mention information and the reference search statement, a scene tag knowledge base is constructed, where the scene tag knowledge base is used to store the corresponding relationships between the scene tag and the core word, the entity mention information and the reference search statement respectively;

[0063] The label determination module is used to determine, for each of the entity word fragments, a target scene label associated with the entity word fragment from the scene label knowledge base to obtain the multiple target scene labels.

[0064] In some embodiments, the result determination module is used to sort the multiple target scene tags based on the first correlation parameter to obtain a plurality of sorted target scene tags; from the multiple search results, the search results linked to the multiple target scene tags sorted before the target number are used as the target search results; or,

[0065] The result determination module is used to use the multiple search results as the target search results.

[0066] In some embodiments, the result determination module is used to send the target search results to the terminal in sequence according to the order of the sorted multiple target scene tags.

[0067] In some embodiments, the plurality of target scene tags include a plurality of levels, each level includes at least one target scene tag, and the target scene tag in the next level belongs to a sub-scene tag of the target scene tag in the previous level; the device further includes:

[0068] The label deletion module is used to delete the target scene labels belonging to the previous level from the target scene labels with a hierarchical relationship, and obtain multiple updated target scene labels.

[0069] In some embodiments, the result determination module is used to determine, for each of the target scene tags, a third correlation parameter between the target scene tag and the search statement; and determine the target search result from the multiple search results based on the first correlation parameter and the third correlation parameter corresponding to each of the target scene tags.

[0070] On the other hand, an embodiment of the present application provides a server, which includes a processor and a memory, wherein the memory stores at least one program code, and the at least one program code is loaded and executed by the processor so that the server implements the information search method described in the above aspects.

[0071] On the other hand, an embodiment of the present application provides a computer-readable storage medium, in which at least one program code is stored. The at least one program code is loaded and executed by a processor to enable a server to implement the information search method described in the above aspects.

[0072] On the other hand, an embodiment of the present application provides a computer program or a computer program product, wherein the computer program or the computer program product stores at least one computer instruction, and the at least one computer instruction is loaded and executed by a processor so that the server implements the information search method described in the above aspects.

[0073] The technical solution provided by the embodiments of the present application brings at least the following beneficial effects:

[0074] In an embodiment of the present application, since the determined target scene label is associated with the entity word fragment, the association between the target scene label and the search statement is closer. On this basis, the target search results to be sent to the terminal are determined in combination with the correlation parameters between each target scene label and the entity word fragment, so that the determined target search results are more in line with the user's search needs, thereby improving the accuracy of information search. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0076] Figure 1 It is a schematic diagram of an implementation environment of an information search method provided in an embodiment of the present application;

[0077] Figure 2 is a flow chart of an information search method provided by an embodiment of the present application;

[0078] Figure 3 It is a flowchart of a method for constructing a scene label knowledge base provided in an embodiment of the present application;

[0079] Figure 4 is a flow chart of an information search method provided by an embodiment of the present application;

[0080] Figure 5 is a flow chart of an information search method provided by an embodiment of the present application;

[0081] Figure 6 is a schematic diagram of an information search method provided in an embodiment of the present application;

[0082] Figure 7 is a structural schematic diagram of an information search device provided in an embodiment of the present application;

[0083] Figure 8 It is a structural diagram of a server provided in an embodiment of the present application. DETAILED DESCRIPTION

[0084] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0085] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0086] The user information involved in the present disclosure may be information authorized by the user or fully authorized by all parties.

[0087] Figure 1 is a schematic diagram of an implementation environment of an information search method provided in an embodiment of the present application, such as Figure 1 As shown, the implementation environment includes: a terminal 101 and a server 102, and the terminal 101 and the server 102 are connected via a wired or wireless network.

[0088] Terminal 101 is used to obtain a search statement input by a user and send the search statement to server 102. Server 102 is used to receive the search statement of terminal 101, perform information search based on the search statement, and send the search result obtained to terminal 101. Accordingly, terminal 101 is also used to receive the search result and display the search result.

[0089] In an embodiment of the present application, a target application provided with background services by server 102 is installed on terminal 101, and terminal 101 can realize functions such as information search and information display through the target application. Optionally, the target application is an application in the operating system of terminal 101, or an application provided by a third party. For example, the target application is a life service application. The user operating terminal 101 enters a search statement in the target application, thereby triggering terminal 101 to send the search statement to server 102.

[0090] In the embodiment of the present application, the terminal 101 is at least one of a mobile phone, a tablet computer, a desktop computer, or a laptop computer. The server 102 can be at least one of a server, a server cluster consisting of multiple servers, a cloud server, a cloud computing platform, and a virtualization center, which is not limited in the embodiment of the present application.

[0091] The information search method provided in the embodiments of the present application can be applied in the following information search scenarios:

[0092] For example, when a user wants to search for a place to take the kids out on weekends, the user enters the search statement "take the kids out on weekends" in terminal 101, and terminal 101 sends the search statement to server 102. Server 102 performs an information search on the search statement and obtains search results, which may be "children's playground", "underwater world" or "amusement park" and other results. Server 102 sends the search results to terminal 101, so that the user can view the search results through terminal 101.

[0093] For another example, when a user wants to search for a park for flying kites, the user enters the search statement "park for flying kites" in terminal 101, and terminal 101 sends the search statement to server 102. Server 102 performs an information search on the search statement to obtain search results. For example, the search result is "People's Park", and server 102 sends the search result to terminal 101, so that the user views the search result through terminal 101.

[0094] It should be noted that the above scenarios are only exemplary descriptions and do not limit the information search scenarios. In addition to being applied to the above scenarios, the present application can also be applied to any other information search scenarios.

[0095] Based on the above implementation environment, the present application embodiment provides an information search method. Figure 2 As an example, the flowchart of an information search method provided in the embodiment of the present application is shown in FIG. Figure 1 The server 102 in the embodiment executes. Figure 2 As shown, the method comprises the following steps:

[0096] In step 201, the server determines a plurality of entity word fragments corresponding to a search statement, wherein at least one entity word fragment among the plurality of entity word fragments includes a plurality of entity words, and the entity word is an entity word in the search statement.

[0097] The search statement is a search statement input by the user through the terminal. In some embodiments, when the user wants to search for a certain life service, the user inputs the search statement through the target application installed in the terminal, so that the terminal sends the search statement to the server, and the server receives the search statement. For example, the search statement is "weekend couple hot springs", "parks for flying kites" or "hotels with hard beds".

[0098] In step 202, the server determines, for each entity word segment, a target scene tag associated with the entity word segment, and obtains a plurality of target scene tags, each of which is linked to a search result.

[0099] Among them, the target scene label is obtained by retrieving entity word fragments from the scene label knowledge base. Each scene label is linked to a search result, which is life service information. The life service information is used to indicate the life services provided by the life service platform corresponding to the target application, that is, the search result is a merchant registered on the life service platform or a product provided by a registered merchant. For example, the search result is a tourist attraction, hotel, restaurant, cinema, food or movie ticket, etc.

[0100] In step 203, the server determines a first correlation parameter between each of the entity word segments and a target scene label associated with the entity word segment.

[0101] Among them, the first correlation parameter is used to indicate the correlation degree between the target scene label and the entity word fragment.

[0102] In step 204, the server determines a target search result from the multiple search results based on the first correlation parameter corresponding to each target scene tag, and sends the target search result to the terminal.

[0103] The terminal receives the target search result sent by the server and displays the target search result so that the user can view the target search result.

[0104] In an embodiment of the present application, since the determined target scene label is associated with the entity word fragment, the association between the target scene label and the search statement is closer. On this basis, the target search results to be sent to the terminal are determined in combination with the correlation parameters between each target scene label and the entity word fragment, so that the determined target search results are more in line with the user's search needs, thereby improving the accuracy of information search.

[0105] In some embodiments, the server constructs a scene tag knowledge base in advance, and the scene tag knowledge base is used to retrieve target scene tags associated with entity word fragments. Figure 3 is a flow chart of a method for constructing a scene tag knowledge base provided in an embodiment of the present application. Figure 3 This embodiment is described by taking the server building a scene tag knowledge base as an example. The method includes the following steps:

[0106] In step 301, the server obtains a plurality of scene tags, where the scene tags are used to represent scene requirements to be searched.

[0107] The terminal has a target application installed, which is a life service application. Merchants are registered in the target application to provide life services to users. The scene tag indicates the scene demand to be searched, that is, the life service provided by the merchant in the target application. For example, the scene tag is bungee jumping, hot spring bathing or swimming.

[0108] In some embodiments, the merchant in the target application provides a variety of life services, so the server divides the various life services according to the service type to obtain multiple vertical fields. For example, the multiple vertical fields are hotel field, tourism field, catering field and other fields. Accordingly, the server independently optimizes the scene tags in each vertical field to obtain multiple scene tags corresponding to the vertical field.

[0109] In step 302, for each scene tag, the server extracts a core word from the scene tag, and based on the core word, determines the entity mention information and reference search statement of the scene tag, where the entity mention information includes at least one entity mention related to the core word, and the reference search statement is a historical search statement related to the core word.

[0110] The server extracts the core words from the scene tags in the following way: the server performs entity recognition on the scene tags to obtain the core words. For example, if the scene tag is hot spring bathing, the core words are hot spring or hot spring bathing; if the scene tag is parent-child travel, the core words are parent-child, travel and play.

[0111] Among them, the entity mention (Mention) of any entity word is the text information that references the entity word. In an embodiment of the present application, the entity mention information of the scene label is the text information that references the core word corresponding to the scene label. For example, the core word is hot spring, and the entity mention information is information such as hot springs, hydrotherapy, and private hot springs. In some embodiments, the scene requirements represented by the scene label are relatively broad, and the server can expand the scene label. Accordingly, the server determines the entity mention information of the scene label based on the core word in the following way: the server expands the core word to obtain the entity mention information.

[0112] The reference search statement is a historical search statement (Query) related to the core word in the historical search process. In some embodiments, the server obtains and stores the search statement input by the user through the terminal. After determining the core word of the scene tag, the server obtains the search statement corresponding to the core word from the stored search statements.

[0113] In step 303, the server constructs a scene label knowledge base based on the scene label, the core word, the entity mention information and the reference search statement. The scene label knowledge base is used to store the correspondence between the scene label and the core word, the entity mention information and the reference search statement.

[0114] In some embodiments, the server constructs a scene label knowledge base in an inverted index manner. Accordingly, the server constructs a core word index based on the correspondence between scene labels and core words; constructs a mention index based on the correspondence between scene labels and entity mention information; and constructs a search statement index based on the correspondence between scene labels and reference search statements, thereby obtaining a scene label knowledge base.

[0115] Optionally, when the entity mention information includes multiple entity mentions, the server sets the correspondence between the scene label and each entity mention in the entity mention information in the mention index, thereby improving the accuracy of the mention index. Optionally, the server segments the reference search statement, and sets the correspondence between the scene label and the segmented reference search statement in the search statement index, thereby improving the accuracy of the search statement index. Optionally, the server segments the scene label, and constructs a single-word index based on the correspondence between the scene label and the segmented single words, thereby expanding the recall to a greater extent.

[0116] In an embodiment of the present application, core words are extracted from scene labels of vertical fields offline, and the core words are expanded to obtain entity mention information and reference search statements corresponding to the core words. Then, the core words, entity mention information and reference search statements of the scene labels are combined to construct a scene label knowledge base, thereby realizing the storage of the correspondence between scene labels and core words, entity mention information and search statements, thereby ensuring relevance while expanding recall, and providing rich data support for the subsequent determination of scene labels.

[0117] In some embodiments, the server performs an information search on the search statement to obtain search results and sends them to the terminal, which displays the search results and the user views the search results through the terminal. The search results triggered by the user can be regarded as search results that the user is interested in, that is, points of interest (POIs). Then the terminal generates a search notification message based on the search results triggered by the user, and reports the search notification message to the server, so that the server can use the search notification information to determine the historical feedback parameters of the scene tag.

[0118] Figure 4 is a flow chart of an information search method provided by an embodiment of the present application. Figure 4 This embodiment is described by taking the server determining the historical feedback parameters of the scene tag as an example. The method includes the following steps:

[0119] In step 401, the server obtains a search notification message reported by any terminal, where the search notification message is used to indicate the feedback status of the terminal on the historical search results.

[0120] The historical search results are search results that have been displayed in the terminal during the historical search process. In some embodiments, the user triggers the terminal to view any search result, the terminal records the triggering operation on any search result, obtains a search notification message, sends the search notification message to the server, and the server receives the search notification message. Optionally, the terminal reports the search notification message in the form of a log.

[0121] It should be noted that the terminal may send the search notification message in real time or at regular intervals, which is not limited in the present application.

[0122] In step 402, the server determines the historical feedback parameter of the scene tag linked to the historical search result based on the search notification message.

[0123] The historical feedback parameter can be represented by the number of historical search results corresponding to the scene tag. The search notification message may indicate multiple historical search results. Accordingly, the implementation of step 402 includes: for any historical search result indicated by the search notification message, the server determines the scene tag linked to the historical search result, the server determines the number of historical search results linked to the scene tag in the multiple historical search results, and takes the sum of the number and the current historical feedback parameter of the scene tag as the updated historical feedback parameter.

[0124] For example, the scene tag is "mountain climbing", and the linked historical search results are "Taishan" and "Xiangshan" respectively, the number corresponding to "Taishan" is 2, and the number corresponding to "Xiangshan" is 5, then the historical feedback parameter of the scene tag "mountain climbing" is 7. For example, taking the user's trigger operation as a click operation as an example, the server performs click feedback statistics on the historical search results in the search notification message, thereby determining the historical feedback parameter P (Query~POI~Tag).

[0125] In some embodiments, since the historical feedback parameter is determined based on the feedback of historical search results, the historical feedback parameter of a scene tag with a higher search frequency is also larger. Optionally, for each scene tag, the server determines the total number of historical search results linked to the scene tag among the multiple historical search results sent to the terminal, and uses the ratio between the number of historical search results linked to the scene tag in the search notification message and the total number as the historical feedback parameter.

[0126] Among them, when the triggering operation is a click operation, the ratio between the number of historical search results linked to the scene tag in the search notification message and the total number is the click-through rate of the scene tag. The server uses the click-through rate of the scene tag as a historical feedback parameter, thereby improving the accuracy of the historical feedback parameter.

[0127] In an embodiment of the present application, the server obtains a large number of search notification messages to determine the historical feedback parameters of the scene tags to improve the accuracy of the historical feedback parameters.

[0128] In an embodiment of the present application, since the user's feedback on the search results can be regarded as the user's feedback on the scene tag linked to the search results, the historical feedback parameters of the scene tag are determined by counting the search notification messages of the terminal, so that the relevance parameters of the scene tag can be adjusted in combination with the historical feedback parameters, thereby providing technical support for improving the accuracy of the relevance parameters.

[0129] Figure 5 is a flow chart of an information search method provided by an embodiment of the present application. Figure 5 This embodiment is described by taking the server performing information search based on a search statement as an example. The method includes the following steps:

[0130] In step 501, the server performs entity recognition on the search statement to obtain multiple entity words.

[0131] The user inputs a search statement through the terminal, the terminal obtains the search statement, and sends the search statement to the server; accordingly, the server obtains the search statement. The entity recognition operation may be named entity recognition (NER).

[0132] Optionally, among the multiple entity words obtained by entity recognition, there may be an entity word used to represent address information, and the entity word has little correlation with the determination of the scene label. Accordingly, after multiple entity words are identified, if there is an entity word used to represent address information among the multiple entity words, the server deletes the entity word used to represent address information from the multiple entity words to obtain an updated entity word, and then performs subsequent processing based on the updated entity word. Among them, the entity word used to represent address information can be a word representing an administrative area (province / city / district / town).

[0133] Optionally, there may be other words in the search statement that are irrelevant to the determination of the scene label, that is, words without actual semantics. For example, the words are "can", "relatively" or "of", etc. Accordingly, the server deletes the words without actual semantics in the search statement to obtain updated entity words, and then performs subsequent processing based on the updated entity words.

[0134] In some embodiments, the server selects some scene tags from multiple scene tags in each vertical field in advance, and forms a whitelist of these scene tags. Optionally, the server forms a whitelist of scene tags with higher search frequencies among multiple scene tags. Accordingly, after performing entity recognition on the search statement, the server matches the entity words obtained by entity recognition with the scene tags in the whitelist. In the case of matching the target scene tag associated with the entity word, the server does not need to combine the identified entity words or search from the scene tag knowledge base, thereby greatly saving operation time and improving operation efficiency. In the case of not matching the target scene tag associated with the entity word, the server performs the operation of step 502.

[0135] It should be noted that after step 501, the server can directly execute the operation of step 502, or first match according to the whitelist and then determine whether to execute the operation of step 502. This application does not limit this.

[0136] In some embodiments, when the search statement carries merchant information, the server does not link the scene tag to such search statement, and directly uses the merchant link linked to the merchant information as the search result, and sends the search result to the terminal. The merchant information is an entity word used to represent merchant component information, such as a merchant name, or the merchant information is merchant intent information. The server performs intent recognition on the search statement to obtain merchant intent information. When the search statement does not carry merchant information, that is, when the multiple entity words obtained by entity recognition do not include entity words used to represent merchants, the server performs the operation of step 502.

[0137] In step 502, the server combines the multiple entity words according to the position of each entity word in the search sentence to obtain the multiple entity word fragments.

[0138] Each entity word has a corresponding position in the search statement. Optionally, the server numbers the entity words in order from front to back in the search statement, and combines multiple entity words according to the sequence numbers obtained by numbering. For example, if the search statement is "park for flying kites", and the multiple entity words identified are "kite" and "park", then the sequence numbers of "kite" and "park" are 1 and 2 respectively.

[0139] Optionally, the server performs permutations and combinations on the multiple entity words to obtain multiple candidate combinations, and each candidate combination includes at least one entity word segment. For each candidate combination, when there is an entity word segment formed by two entity words with non-adjacent positions in the candidate combination, that is, an entity word segment formed by two entity words with discontinuous serial numbers, the server deletes the candidate combination to obtain an updated candidate combination, and the server uses the entity word segments included in the updated candidate combination as the multiple entity word segments and performs the operation in step 503.

[0140] For example, if the multiple entity words are T1, T2, and T3 respectively, then the sequence combinations are (T1T2, T3), (T1T2T3), (T1, T2, T3), (T1, T2T3), and T1T2 is called an entity word segment.

[0141] It should be noted that the entity words obtained by entity recognition may not be the original words in the search statement. Correspondingly, in this case, when the server determines the position of the entity word in the search statement, it uses the position of the word in the search statement corresponding to the entity word as the position of the entity word. For example, the search statement is "couple private hot spring", and the entity words obtained by entity recognition are "couple", "private", and "hot spring", and the server uses the position where "private" is located as the position of "private".

[0142] In the embodiment of the present application, since in the information search scenario with a relatively wide search scope, the number of entity words recognized from the search statement is large, and there may be a connection between adjacent two entity words, the multiple entity words are sequence-combined according to the positions of the entity words in the search statement to obtain entity word segments. In this way, the accuracy of the scenario label determined according to the entity word segments is relatively high.

[0143] In step 503, for each of the entity word segments, the server determines the target scenario label associated with the entity word segment to obtain multiple target scenario labels, and each of the target scenario labels is linked to a search result.

[0144] Among them, the server has previously constructed a scenario label knowledge base, so the server can determine the target scenario label in combination with the scenario label knowledge base. Optionally, the implementation manner of step 503 includes: for each of the entity word segments, the server determines the target scenario label associated with the entity word segment from the scenario label knowledge base to obtain the multiple target scenario labels.

[0145] Among them, the server searches the scene label knowledge base and recalls the target scene label. Since the scene label knowledge base may include multiple indexes, each index has its own priority, then for each entity word fragment, the server retrieves the entity word fragment from each index in turn according to the priority of the index. In the embodiment of the present application, the multiple indexes included in the scene label knowledge base are taken as core word index, mention index, search statement index and single word index as examples for explanation.

[0146] Among them, the priority of multiple indexes can be set as needed, and the embodiments of the present application do not limit this. For example, the server determines in advance that the priorities of multiple indexes are core word index, mention index, search statement index and single word index, that is, the core word index has the highest priority and the single word index has the lowest priority. For example, the priorities of the core word index, mention index, search statement index and single word index are 4, 3, 2 and 1 respectively, where the larger the value, the higher the priority. Correspondingly, for each entity word fragment, the server retrieves the target scene label associated with the entity word fragment from the core word index, mention index, search statement index and single word index in turn.

[0147] It should be noted that the server can select any index from a variety of indexes and perform retrieval based on the index, thereby saving retrieval time and improving retrieval efficiency. For example, the server performs retrieval from a core word index or a mention index.

[0148] In some embodiments, the server first performs the operations of steps 504-505 to determine the first correlation parameter, and then performs the operation of step 506 to determine the third correlation parameter. Alternatively, the server first performs the operation of step 506 and then performs the operations of steps 504-505. Alternatively, the server performs the operations of steps 504-505 and 506 at the same time. This application does not limit this.

[0149] In step 504, for each entity word fragment, feature extraction is performed on the entity word fragment to obtain a first feature, feature extraction is performed on the target scene label to obtain a second feature, and a second correlation parameter between the first feature and the second feature is determined.

[0150] The first feature may be a vector representation (Tagvec) of an entity word fragment, and the second feature may be a vector representation of a target scene tag (Tag). Optionally, the server determines the second correlation parameter between the first feature and the second feature in the following manner: the server determines the cosine similarity between the first feature and the second feature, and determines the second correlation parameter based on the cosine similarity, wherein the cosine similarity is positively correlated with the second correlation parameter.

[0151] Among them, the larger the cosine similarity, the more similar the first feature and the second feature are, that is, the closer the target scene label is to the entity word fragment. Optionally, the server converts the cosine similarity according to a preset ratio to obtain a second correlation parameter, so as to facilitate subsequent adjustment of the second correlation parameter. The preset ratio is a pre-set ratio, and the preset ratio can be set as needed, which is not limited in the embodiments of the present application. For example, the preset ratio is 1:10, the cosine similarity is 0.1, and the second correlation parameter is 0.1×10=1. For another example, the preset ratio is 1:100, the cosine similarity is 0.2, and the second correlation parameter is 0.2×100=20.

[0152] In step 505, the server adjusts the second correlation parameter based on the priority and historical feedback parameters corresponding to the target scene tag to obtain the first correlation parameter.

[0153] Among them, the historical feedback parameter is used to represent the feedback status of the target scene label during the historical search process, the priority is the priority of the target retrieval method, and the target retrieval method is the retrieval method used to retrieve the target scene label.

[0154] For example, the target scene label is determined based on the core word index, and the priority of the core word index is 1, then the priority corresponding to the target scene label is 1.

[0155] In some embodiments, the larger the historical feedback parameter is, the more feedback the user has given to the target scene tag, and the higher the priority is, the stronger the association between the target scene tag and the search statement is, thereby indicating that the target scene tag is more in line with the user's scene requirements, and thus the second relevance parameter is more referenceable. Accordingly, the implementation method of step 505 includes: the server determines a first increment of the second relevance parameter based on a first difference between the historical feedback parameter and the feedback threshold, and the first difference is positively correlated with the first increment; the server determines a second increment of the second relevance parameter based on a second difference between the priority and the priority threshold, and the second difference is positively correlated with the second increment; the server adjusts the second relevance parameter based on the first increment and the second increment to obtain the first relevance parameter.

[0156] The feedback threshold can be set as needed, and the embodiment of the present application does not limit this. For example, the feedback threshold is set to 0. The priority threshold is set as needed, and the embodiment of the present application does not limit this. For example, the priority threshold is set to 0.

[0157] Optionally, the server adjusts the second correlation parameter based on the first value-added and the second value-added to obtain the first correlation parameter in the following manner: the server uses the sum of the first value-added, the second value-added and the second correlation parameter as the first correlation parameter.

[0158] In the embodiment of the present application, on the one hand, since the historical feedback parameter represents the historical feedback on the target scene label, the larger the first difference value determined according to the historical feedback parameter, the larger the historical feedback parameter of the target scene label, and thus the larger the first increment, the greater the reference of the second correlation parameter, and the first correlation parameter is more accurate. On the other hand, since the priority corresponding to the target scene label represents the degree of association between the target scene label and the search statement, the larger the second difference value determined according to the priority, the stronger the association between the target scene label and the target scene label and the search statement, and thus the larger the second increment, the greater the reference of the second correlation parameter, and the first correlation parameter is more accurate.

[0159] In some other embodiments, the server determines the first correlation parameter based on the historical feedback parameter of the target scene tag and the second correlation parameter. Accordingly, the implementation of step 505 is replaced by: the server adjusts the second correlation parameter based on the historical feedback parameter corresponding to the target scene tag to obtain the first correlation parameter.

[0160] Optionally, the server determines a first value-added, and uses the sum of the first value-added and the second correlation parameter as the first correlation parameter. The first value-added is determined in the same manner as the first value-added in the above embodiment, and will not be described in detail here.

[0161] In an embodiment of the present application, since the historical feedback parameter can represent the user's historical feedback on the target scene label, the first correlation parameter of the target scene label is determined by combining the historical feedback parameter of the target scene label, so that the determined first correlation parameter has higher accuracy.

[0162] In some other embodiments, the server determines the first correlation parameter according to the priority corresponding to the target scene tag and the second correlation parameter. Accordingly, the implementation of step 505 is replaced by: the server adjusts the second correlation parameter based on the priority corresponding to the target scene tag to obtain the first correlation parameter.

[0163] Optionally, the server determines a second value-added value, and uses the sum of the second value-added value and the second correlation parameter as the first correlation parameter. The second value-added value is determined in the same manner as the second value-added value in the above embodiment, and will not be described in detail here.

[0164] In an embodiment of the present application, since the priority corresponding to the target scene tag represents the degree of association between the target scene tag and the search statement, the first relevance parameter of the target scene tag is determined by combining the priority corresponding to the target scene tag, so that the accuracy of the determined first relevance parameter is higher.

[0165] It should be noted that the server can adjust the second correlation parameter in combination with the historical feedback parameter, and can also adjust the second correlation parameter in combination with the priority. Alternatively, the server can also adjust the second correlation parameter in combination with the historical feedback parameter and the priority. This is not limited in the embodiments of the present application.

[0166] In step 506, the server determines, for each target scene tag, a third correlation parameter between the target scene tag and the search statement.

[0167] In some embodiments, the implementation of step 506 is similar to the implementation of steps 504-505, and will not be repeated here.

[0168] It should be noted that the search statement may be a search statement after the entity words used to represent the address information have been deleted. In the embodiment of the present application, after obtaining the first correlation parameter P(Tag~e) between the target scene tag (Tag) and the entity word fragment (e), and the third correlation parameter P(Tag~Query) between the target scene tag and the search statement (Query), the server selects the target search result from the multiple search results based on the first correlation parameter and the third correlation parameter of each target scene tag, that is, performs the operation of step 507.

[0169] In step 507, the server determines the target search result from the multiple search results based on the first relevance parameter and the third relevance parameter corresponding to each target scene tag.

[0170] Among them, for the target scene tags with relatively high values ​​of the first correlation parameter and the third correlation parameter, they are more closely associated with the search statement and are more in line with the user's search needs. Accordingly, the implementation method of step 507 includes: the server filters out the target scene tags whose first correlation parameter and second correlation parameter are not less than the correlation threshold from multiple target scene tags, and uses the search result corresponding to the filtered target scene tag as the target search result.

[0171] Among them, the relevance threshold can be set as needed, and the embodiment of the present application does not limit this. In the embodiment of the present application, the server filters out the target scene tags whose first relevance parameter and the third relevance parameter are less than the relevance threshold, so that the determined target scene tags have a high relevance to the entity word fragment and the search statement, and thus the accuracy of the target scene tags is high.

[0172] In some embodiments, the server can determine the target search results in combination with the first relevance parameter to save screening time. Accordingly, steps 506-507 are replaced by: the server determines the target search results from multiple search results based on the first relevance parameter corresponding to each target scene tag.

[0173] The server may select target scene tags whose first relevance parameter is not less than the relevance threshold from the multiple target scene tags, and use the search results corresponding to the selected target scene tags as the target search results. Alternatively, the server may sort the multiple target scene tags based on the first relevance parameter to obtain multiple sorted target scene tags; from the multiple search results, the search results linked to the multiple target scene tags sorted before the target number are used as the target search results.

[0174] The target number of bits can be set as needed, and this embodiment of the present application does not limit this.

[0175] In an embodiment of the present application, by first sorting multiple target scene tags and then selecting the target scene tags ranked before the target number, the determined target search results are more closely matched with the user's search needs, and the number of search results is reduced, thereby reducing the amount of data transmitted.

[0176] In other embodiments, the server uses multiple search results as the target search results. Since each search result is a search result linked to a target scenario tag, and the search result is related to the user's search needs represented by the search statement to a certain extent, the multiple search results can be directly used as the target search results to save screening time and improve the efficiency of information search.

[0177] Among them, multiple target scene tags may include multiple levels, each level contains at least one target scene tag, and the target scene tag in the next level belongs to the sub-scene tag of the target scene tag in the previous level. For example, taking the tourism field as an example, the scene tags are hot spring bathing, couple hot spring bathing and private hot spring bathing, among which couple hot spring bathing and private hot spring bathing are both sub-scene tags of hot spring bathing.

[0178] Optionally, for each vertical field, the server determines in advance the hierarchical relationship between multiple scene tags in the vertical field, obtains multiple levels of scene tags, and stores the hierarchical relationship between multiple scene tags. For example, the server stores the hierarchical relationship between multiple scene tags in the form of a knowledge graph. After obtaining multiple target scene tags, the server determines whether the multiple target scene tags include target scene tags with a hierarchical relationship based on the stored multiple levels of scene tags.

[0179] Accordingly, after determining multiple target scene tags, that is, before step 504, the information search method provided in the embodiment of the present application also includes the following steps: the server deletes the target scene tags belonging to the previous level from the target scene tags with a hierarchical relationship, and obtains multiple updated target scene tags.

[0180] After deleting the target scene tags belonging to the previous level, the server performs operations of steps 504-507 according to the remaining target scene tags.

[0181] In an embodiment of the present application, by combining the hierarchical relationship in the target scene label, the main and sub scenes and redundant scenes are merged and selected, thereby achieving joint disambiguation between scenes, thereby improving the accuracy of label determination.

[0182] It should be noted that, since the target application is a life service application, the information search scenarios applicable to the embodiments of the present application are mostly information search scenarios that are strongly related to addresses, and the server can also filter multiple target scene tags in combination with the address information. Accordingly, the server selects a target scene tag corresponding to the address information from multiple target scene tags based on the address information of the terminal, and then performs the operations of steps 504-507 based on the selected target scene tag.

[0183] The implementation method of the server determining the address information of the terminal is as follows: when the entity words obtained by entity recognition include entity words used to represent address information, the server uses the address information as the address information of the terminal; when the entity words obtained by entity recognition do not include entity words used to represent address information, the server sends an address acquisition request to the terminal, and the terminal responds to the address acquisition request by sending its own address information to the server, and the server receives the address information. Alternatively, the terminal may also send its own address information together with the search statement sent to the server.

[0184] In step 508, the server sends the target search result to the terminal.

[0185] Among them, the target search result has at least one of the corresponding first relevance parameter or second relevance parameter, and the larger the relevance parameter, the more the target search result meets the user's search needs. Accordingly, the server sends the target search result to the terminal in sequence according to the order of the sorted multiple target scene tags. Among them, the server can sort multiple target scene tags according to the first relevance parameter, or sort multiple target scene tags according to the second relevance parameter, and the embodiment of the present application does not limit this. Alternatively, the server packages the determined target search results to obtain a search result package, and sends the search result package to the terminal, thereby reducing the data interaction between the server and the terminal and improving the sending efficiency. Among them, each target search result in the search result package is marked with a serial number. Optionally, the server marks each target search result with a serial number in advance according to the sorting result, and the serial number corresponding to each target search result is the number of digits of the target scene label corresponding to the target search result in the sorting result.

[0186] The terminal receives the target search result and displays the target search result. Optionally, the terminal receives a search result package and displays the target search result in sequence in a display interface according to the sequence number corresponding to each target search result in the search result package.

[0187] In the embodiment of the present application, in the information search scenario with a relatively wide search scope, that is, in the pan-scenario search, the above scheme can analyze the core search requirements in the search statement and link them to the existing scene tag knowledge system, thereby guiding the recall of more accurate and more relevant search results. After testing by technicians in the field, the information search method provided by the embodiment of the present application has a recall rate of up to 84% and an accuracy rate of up to 93% in the information search process in the tourism field.

[0188] In the embodiments of the present application, on the one hand, the scene search is analyzed according to the characteristics of scene search in vertical fields, and split into multi-granularity inverted indexes to obtain a scene label knowledge base, thereby effectively improving the recall coverage of target scene labels. On the other hand, the recalled multiple target scene labels are used as a pan-scene set, combined with the historical feedback parameters obtained by aggregating the user's search behavior for merchants, to determine the relevance parameters of each target scene label in the sorting link, and at the same time, a variety of judgment methods are combined to disambiguate the relationship between multiple scenes such as composite scenes, main and secondary scenes, etc., effectively improving the accuracy of information search.

[0189] For example, see Figure 6, the search statement (Query) includes address information (City), the search statement is segmented and entity recognized, the entity word corresponding to the address information is deleted from the recognized entity words (X recognition result), and the remaining entity words are fully matched in the whitelist. If there is a matching scene tag, the matched scene tag is used as the target scene tag (Tag). If there is no matching scene tag, the remaining entity words are combined to obtain entity word fragments, and then a reverse search is performed in the scene tag knowledge base to recall the target scene tag. The obtained target scene tags are sorted, for example, from large to small according to the first relevance parameter or the third relevance parameter. In addition, the corresponding relationship between the address information corresponding to the target scene tag and the target scene tag is also referred to, and the sorted target scene tags are finally judged to obtain the target search results corresponding to the filtered target scene tags.

[0190] In an embodiment of the present application, since the determined target scene label is associated with the entity word fragment, the association between the target scene label and the search statement is closer. On this basis, the target search results to be sent to the terminal are determined in combination with the correlation parameters between each target scene label and the entity word fragment, so that the determined target search results are more in line with the user's search needs, thereby improving the accuracy of information search.

[0191] Figure 7 FIG. 1 is a schematic diagram of the structure of an information search device provided in an embodiment of the present application. Figure 7 As shown, the device comprises:

[0192] A fragment determination module 701 is used to determine a plurality of entity word fragments corresponding to a search sentence, wherein at least one entity word fragment among the plurality of entity word fragments includes a plurality of entity words, and the entity word is an entity word in the search sentence;

[0193] A tag determination module 702 is used to determine, for each entity word segment, a target scene tag associated with the entity word segment, to obtain a plurality of target scene tags, each of which is linked to a search result;

[0194] A first parameter determination module 703, used to determine a first correlation parameter between each entity word segment and a target scene label associated with the entity word segment;

[0195] The result determination module 704 is used to determine a target search result from multiple search results based on the first correlation parameter corresponding to each target scene tag, and send the target search result to the terminal.

[0196] In some embodiments, the fragment determination module 701 is used to perform entity recognition on the search sentence to obtain multiple entity words; according to the position of each entity word in the search sentence, the multiple entity words are combined to obtain the multiple entity word fragments.

[0197] In some embodiments, the first parameter determination module 703 includes:

[0198] A feature extraction unit is used to perform feature extraction on each entity word segment to obtain a first feature, perform feature extraction on the target scene label to obtain a second feature, and determine a second correlation parameter between the first feature and the second feature;

[0199] The parameter adjustment unit is used to adjust the second correlation parameter based on the historical feedback parameter of the target scene label to obtain the first correlation parameter, and the historical feedback parameter is used to represent the feedback status of the target scene label in the historical search process.

[0200] In some embodiments, the parameter adjustment unit is used to determine a first increased value of the second correlation parameter based on a first difference between the historical feedback parameter and a feedback threshold, and the first difference is positively correlated with the first increased value; and the sum of the first increased value and the second correlation parameter is used as the first correlation parameter.

[0201] In some embodiments, the apparatus further comprises:

[0202] The second parameter determination module is used to determine the historical feedback parameter of the scene tag linked to the historical search results based on the search notification message reported by any terminal, and the search notification message is used to indicate the feedback of the terminal to the historical search results.

[0203] In some embodiments, the first parameter determination module 703 includes:

[0204] A feature extraction unit is used to perform feature extraction on each entity word segment to obtain a first feature, perform feature extraction on the target scene label to obtain a second feature, and determine a second correlation parameter between the first feature and the second feature;

[0205] The parameter adjustment unit is also used to adjust the second correlation parameter based on the priority corresponding to the target scene label to obtain the first correlation parameter. The priority is the priority of the target retrieval method, and the target retrieval method is the retrieval method used to retrieve the target scene label.

[0206] In some embodiments, the parameter adjustment unit is further used to determine a second increased value of the second correlation parameter based on a second difference between the priority and the priority threshold, the second difference being positively correlated with the second increased value; and to use the sum of the second increased value and the second correlation parameter as the first correlation parameter.

[0207] In some embodiments, the apparatus further comprises:

[0208] A knowledge base construction module is used to obtain multiple scene tags, where the scene tags are used to represent the scene requirements to be searched; for each of the scene tags, a core word is extracted from the scene tag, and based on the core word, entity mention information and a reference search statement of the scene tag are determined, where the entity mention information includes at least one entity mention related to the core word, and the reference search statement is a historical search statement related to the core word; based on the scene tag, the core word, the entity mention information and the reference search statement, a scene tag knowledge base is constructed, where the scene tag knowledge base is used to store the corresponding relationships between the scene tag and the core word, the entity mention information and the reference search statement respectively;

[0209] The label determination module 702 is used to determine, for each entity word segment, a target scene label associated with the entity word segment from the scene label knowledge base to obtain the multiple target scene labels.

[0210] In some embodiments, the result determination module 704 is used to sort the multiple target scene tags based on the first correlation parameter to obtain a plurality of sorted target scene tags; from the multiple search results, the search results linked to the multiple target scene tags sorted before the target number are used as the target search results; or,

[0211] The result determination module 704 is configured to use the multiple search results as the target search result.

[0212] In some embodiments, the result determination module 704 is used to send the target search results to the terminal in sequence according to the sorted order of the multiple target scene tags.

[0213] In some embodiments, the plurality of target scene tags include a plurality of levels, each level includes at least one target scene tag, and the target scene tag in the next level is a sub-scene tag of the target scene tag in the previous level; the device further includes:

[0214] The label deletion module is used to delete the target scene labels belonging to the previous level from the target scene labels with a hierarchical relationship, and obtain multiple updated target scene labels.

[0215] In some embodiments, the result determination module 704 is used to determine, for each target scene tag, a third correlation parameter between the target scene tag and the search statement; and determine the target search result from the multiple search results based on the first correlation parameter and the third correlation parameter corresponding to each target scene tag.

[0216] In an embodiment of the present application, since the determined target scene label is associated with the entity word fragment, the association between the target scene label and the search statement is closer. On this basis, the target search results to be sent to the terminal are determined in combination with the correlation parameters between each target scene label and the entity word fragment, so that the determined target search results are more in line with the user's search needs, thereby improving the accuracy of information search.

[0217] It should be understood that the above Figure 7 When the device provided realizes its functions, only the division of the above-mentioned functional modules is used as an example for illustration. In actual applications, the above-mentioned functions are assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0218] Figure 8 It is a structural diagram of a server provided in an embodiment of the present application. The server 800 may have relatively large differences due to different configurations or performances, and includes one or more processors (Central Processing Units, CPU) 801 and one or more memories 802, wherein the one or more memories 802 store at least one program code, and the at least one program code is loaded and executed by the one or more processors 801 to implement the information search method in the above embodiment. Of course, the server 800 also has components such as a wired or wireless network interface, a keyboard, and an input and output interface for input and output. The server 800 also includes other components for implementing device functions, which will not be repeated here.

[0219] In an exemplary embodiment, a computer-readable storage medium is further provided, in which at least one program code is stored. The at least one program code is loaded and executed by a processor to enable a server to implement the information search method in the above embodiment.

[0220] Optionally, the computer-readable storage medium is a read-only memory (ROM), a random access memory (RAM), a compact disc (CD-ROM), a magnetic tape, a floppy disk, an optical data storage device, etc.

[0221] In an exemplary embodiment, a computer program or a computer program product is also provided, in which at least one computer instruction is stored, and the at least one computer instruction is loaded and executed by a processor to enable a server to implement the information search method in the above embodiment.

[0222] The above are merely exemplary embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. An information search method, characterized in that: The method comprises: Determine multiple entity word fragments corresponding to the search sentence, wherein at least one entity word fragment among the multiple entity word fragments includes multiple entity words, and the entity words are entity words in the search sentence; For each entity word fragment, determining a target scene label associated with the entity word fragment, and obtaining a plurality of target scene labels, each of which is linked to a search result; Determine a first correlation parameter between each of the entity word fragments and a target scene tag associated with the entity word fragment; determine a target search result from a plurality of search results based on the first correlation parameter corresponding to each of the target scene tags, and send the target search result to a terminal; the plurality of target scene tags include a plurality of levels, each level includes at least one target scene tag, and the target scene tag in the next level belongs to a sub-scene tag of the target scene tag in the previous level; Before determining a target search result from a plurality of search results based on the first correlation parameter corresponding to each of the target scene tags, the method further includes: From the target scene labels that have a hierarchical relationship, the target scene labels belonging to the previous level are deleted to obtain multiple updated target scene labels.

2. The method according to claim 1, characterized in that The step of determining a plurality of entity word fragments corresponding to the search statement includes: Performing entity recognition on the search sentence to obtain multiple entity words; The multiple entity words are combined according to the position of each entity word in the search sentence to obtain the multiple entity word fragments.

3. The method according to claim 1, characterized in that The determining of a first correlation parameter between each of the entity word segments and a target scene label associated with the entity word segment includes: For each of the entity word fragments, feature extraction is performed on the entity word fragment to obtain a first feature, feature extraction is performed on the target scene label to obtain a second feature, and a second correlation parameter between the first feature and the second feature is determined; Based on the historical feedback parameter of the target scene label, the second correlation parameter is adjusted to obtain the first correlation parameter, and the historical feedback parameter is used to represent the feedback status of the target scene label in the historical search process.

4. The method according to claim 3, characterized in that The adjusting the second correlation parameter based on the historical feedback parameter of the target scene label to obtain the first correlation parameter includes: determining a first increment of the second correlation parameter based on a first difference between the historical feedback parameter and a feedback threshold, wherein the first difference is positively correlated with the first increment; The sum of the first increment and the second correlation parameter is taken as the first correlation parameter.

5. The method according to claim 3, characterized in that: The method further comprises: Based on a search notification message reported by any terminal, a historical feedback parameter of a scene tag linked to a historical search result is determined, wherein the search notification message is used to indicate feedback status of the terminal on the historical search result.

6. The method according to claim 1, characterized in that The determining of a first correlation parameter between each of the entity word segments and a target scene label associated with the entity word segment includes: For each of the entity word fragments, feature extraction is performed on the entity word fragment to obtain a first feature, feature extraction is performed on the target scene label to obtain a second feature, and a second correlation parameter between the first feature and the second feature is determined; Based on the priority corresponding to the target scene tag, the second correlation parameter is adjusted to obtain the first correlation parameter, where the priority is the priority of the target retrieval method, and the target retrieval method is the retrieval method used to retrieve the target scene tag.

7. The method according to claim 6, characterized in that The adjusting the second correlation parameter based on the priority corresponding to the target scene label to obtain the first correlation parameter includes: determining a second increment of the second relevance parameter based on a second difference between the priority and the priority threshold, the second difference being positively correlated with the second increment; The sum of the second increment and the second correlation parameter is used as the first correlation parameter.

8. The method according to claim 1, characterized in that The method further comprises: Acquire multiple scene tags, where the scene tags are used to represent scene requirements to be searched; For each of the scene tags, extract a core word from the scene tag, and determine entity mention information and a reference search statement of the scene tag based on the core word, wherein the entity mention information includes at least one entity mention related to the core word, and the reference search statement is a historical search statement related to the core word; Based on the scene label, the core word, the entity mention information and the reference search sentence, a scene label knowledge base is constructed, wherein the scene label knowledge base is used to store the corresponding relationships between the scene label and the core word, the entity mention information and the reference search sentence respectively; For each entity word segment, determining a target scene label associated with the entity word segment to obtain multiple target scene labels includes: For each of the entity word fragments, a target scene label associated with the entity word fragment is determined from the scene label knowledge base to obtain the multiple target scene labels.

9. The method according to claim 1, characterized in that: The step of determining a target search result from a plurality of search results based on the first correlation parameter corresponding to each of the target scene tags includes: Based on the first correlation parameter, the multiple target scene tags are sorted to obtain a plurality of sorted target scene tags; from the multiple search results, the search results linked to the multiple target scene tags sorted before the target number are used as the target search results; or, the multiple search results are used as the target search results.

10. The method according to claim 1, characterized in that The sending the target search results to the terminal includes: sending the target search results to the terminal in sequence according to the sorted order of the plurality of target scene tags.

11. The method according to claim 1, characterized in that: The step of determining a target search result from a plurality of search results based on the first correlation parameter corresponding to each of the target scene tags includes: For each of the target scene tags, determining a third correlation parameter between the target scene tag and the search statement; The target search result is determined from the multiple search results based on the first relevance parameter and the third relevance parameter corresponding to each of the target scene tags.

12. An information search device, characterized in that: The device comprises: A fragment determination module, used to determine a plurality of entity word fragments corresponding to a search sentence, wherein at least one entity word fragment among the plurality of entity word fragments includes a plurality of entity words, and the entity words are entity words in the search sentence; A tag determination module, used for determining, for each entity word segment, a target scene tag associated with the entity word segment, to obtain a plurality of target scene tags, each of which is linked to a search result; A parameter determination module, used to determine a first correlation parameter between each of the entity word fragments and a target scene label associated with the entity word fragment; A result determination module, configured to determine a target search result from multiple search results based on a first correlation parameter corresponding to each target scene tag, and send the target search result to a terminal; The multiple target scene tags include multiple levels, each level includes at least one target scene tag, and the target scene tag in the next level is a sub-scene tag of the target scene tag in the previous level; Before determining a target search result from a plurality of search results based on the first correlation parameter corresponding to each of the target scene tags, the method further includes: From the target scene labels that have a hierarchical relationship, the target scene labels belonging to the previous level are deleted to obtain multiple updated target scene labels.

13. A server, characterized in that: The server includes a processor and a memory, wherein at least one program code is stored in the memory, and the at least one program code is loaded and executed by the processor so that the server implements the information search method according to any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one program code, and the at least one program code is loaded and executed by the processor so that the server implements the information search method according to any one of claims 1 to 11.

Images