WO2023143640A1 - 搜索意图的查询理解方法和装置、存储介质和电子设备 - Google Patents

搜索意图的查询理解方法和装置、存储介质和电子设备 Download PDF

Info

Publication number
WO2023143640A1
WO2023143640A1 PCT/CN2023/084548 CN2023084548W WO2023143640A1 WO 2023143640 A1 WO2023143640 A1 WO 2023143640A1 CN 2023084548 W CN2023084548 W CN 2023084548W WO 2023143640 A1 WO2023143640 A1 WO 2023143640A1
Authority
WO
WIPO (PCT)
Prior art keywords
information
recall
entity
searched
core
Prior art date
Application number
PCT/CN2023/084548
Other languages
English (en)
French (fr)
Other versions
WO2023143640A8 (zh
WO2023143640A9 (zh
Inventor
许林隆
Original Assignee
浙江口碑网络技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 浙江口碑网络技术有限公司 filed Critical 浙江口碑网络技术有限公司
Publication of WO2023143640A1 publication Critical patent/WO2023143640A1/zh
Publication of WO2023143640A9 publication Critical patent/WO2023143640A9/zh
Publication of WO2023143640A8 publication Critical patent/WO2023143640A8/zh

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
    • G06F16/367—Ontology

Definitions

  • the present application relates to the technical field of computer applications, and in particular to a query comprehension method and device for search intent.
  • the present application relates to a computer storage medium and an electronic device at the same time.
  • keywords of the information to be searched can be input through a search engine set on the network platform, and then information related to the information to be searched can be obtained.
  • the present application provides a query comprehension method for search intent to solve the technical problems of incomplete understanding of search intent and poor recall effect in the prior art.
  • the present application provides a query comprehension method for search intentions, including: performing entity recognition on the information to be searched, determining the entity type information in the information to be searched; according to the core entity and attribute entity in the entity type information, and the knowledge map According to the association relationship between, obtain the core information of the core entity and the attribute information of the attribute entity; according to the entity type information, determine the recall domain of the information to be searched; according to the core information and the attribute information, Determining the recall content of the information to be searched; generating a recall condition of the information to be searched according to the recall domain and the recall content.
  • performing entity identification on the information to be searched, and determining entity type information in the information to be searched includes: determining whether there is a nested relationship between the entity types of the information to be searched for entity identification; if If there is a nested relationship among the entity types for which entity identification is to be performed on the information to be searched, then the entities corresponding to the entity types with the nested relationship are regarded as entities of the same type and determined as the entity type information.
  • the acquiring the core information of the core entity and the attribute information of the attribute entity according to the association relationship between the core entity and the attribute entity in the entity type information and the knowledge graph includes: establishing An association relationship between the core entity and the attribute entity and the entity type corresponding to the knowledge map; according to the association relationship, the core information of the core entity and the attribute information of the attribute entity are obtained.
  • generating the recall condition of the information to be searched according to the recall domain and the recall content includes: generating a knowledge type recall of the information to be searched according to the recall domain and the recall content condition.
  • the recall of the information to be searched is determined according to the core information and attribute information
  • the content includes: according to the knowledge graph, performing field granularity analysis on the core information and the attribute information as a whole, and obtaining unit fields used to describe the core information and the attribute information; according to the knowledge graph, Obtaining the rewritten field of the unit field; determining the rewritten field as the recall content of the key field in the recall condition; generating the recall condition of the information to be searched according to the recall field and the recall content, including : Generate a key field type recall condition of the information to be searched according to the recall domain and the key field recall content.
  • the determining the recall content of the information to be searched according to the core information and attribute information includes: performing field granularity on the core information and the attribute information as a whole according to the knowledge map Analyzing, obtaining the unit field used to describe the core information and the attribute information; according to the unit field, determining the weight and/or compactness of the unit field; according to the weight and/or compactness, Determining the recall content of the information to be searched; generating the recall condition of the information to be searched according to the recall domain and the recall content includes: according to the recall domain and according to the weight and/or closeness The determined recall content generates a key field type recall condition of the information to be searched.
  • the determining the recall content of the information to be searched according to the weight and/or closeness includes: discarding the unit field according to the weight and/or closeness of the unit field Field and/or rewriting, obtaining the target unit field; determining the target unit field as the recall content; generating the recall condition of the information to be searched according to the recall domain and the recall content includes: according to the The recall domain and the target unit field are used to generate a key field type recall condition of the information to be searched.
  • it further includes: when the entity type information includes at least one entity of a subject entity, a scene entity, and a category entity, performing labeling processing on the entity type information; according to the labeling process , predicting label information corresponding to the at least one type of entity.
  • determining the recall domain of the information to be searched according to the entity type information; determining the recall content of the information to be searched according to the core information and the attribute information includes: The subject entity in the entity type information is determined as the subject recall domain; the subject tag of the subject entity is determined as the subject tag recall content; and the information to be searched is generated according to the recall domain and the recall content
  • the recall condition includes: generating a hashtag type recall condition of the information to be searched according to the topic recall field and the hashtag recall content.
  • determining the recall domain of the information to be searched according to the entity type information; determining the recall content of the information to be searched according to the core information and the attribute information includes: The scene entity in the entity type information is determined as a scene recall field; the scene label of the scene entity is determined as the scene label recall content; the generating of the information to be searched is based on the recall field and the recall content
  • the recall condition includes: generating the scene label class of the information to be searched according to the scene recall field and the scene label recall content Type recall conditions.
  • determining the recall domain of the information to be searched according to the entity type information; determining the recall content of the information to be searched according to the core information and the attribute information includes: The category entity in the entity type information is determined as the category recall domain; the category label of the category entity is determined as the category label recall content; the generation of the recall content according to the recall domain and the recall content
  • the recall condition of the information to be searched includes: generating the category label type recall condition of the information to be searched according to the category recall field and the category label recall content.
  • it also includes: determining the industry type of the information to be searched; performing entity identification on the information to be searched, and determining the entity type information in the information to be searched, including: within the scope of the industry type Perform entity identification on the information to be searched, and determine the entity type information in the information to be searched.
  • it also includes: performing error correction processing on the information to be searched; performing entity identification on the information to be searched, and determining the entity type information in the information to be searched, including: according to the error correction processing of the Perform entity identification on the information to be searched, and determine the entity type information in the information to be searched.
  • it also includes: when the entity type information includes address entity type information, determining the address field in the address entity type information as an address recall domain; normalizing the address field or the address field The converted standard address name is determined as the address recall content; the recall condition for generating the information to be searched according to the recall domain and the recall content includes: generating according to the address recall domain and the address recall content The recall condition of the address type of the information to be searched.
  • the present application also provides a query comprehension device for search intentions, including: a first determination unit, configured to perform entity identification on the information to be searched, and determine entity type information in the information to be searched; a second determination unit, configured to According to the relationship between the core entity and the attribute entity in the entity type information and the knowledge graph, the core information of the core entity and the attribute information of the attribute entity are obtained; the third determination unit is used to obtain the core information of the core entity and the attribute information of the attribute entity; the third determining unit is used to obtain the entity type information according to the entity type information , determining the recall domain of the information to be searched; determining the recall content of the information to be searched according to the core information and the attribute information; a generating unit configured to generate the recall domain and the recall content according to the recall domain The recall criteria for the information to be searched.
  • the present application also provides a computer storage medium for storing computer program instructions; when the computer program instructions are read and executed by a processor, the steps of the query comprehension method for the above-mentioned search intent are executed.
  • the present application also provides an electronic device, including: a processor; a memory for storing a program of instructions executable by the processor, and when the program is read and executed by the processor, it executes the query comprehension method of the above-mentioned search intent A step of.
  • the present application provides a query comprehension method for search intentions, which obtains the core information of the core entities and all of them by identifying the relationship between the core entities and attribute entities in the entity type information in the information to be searched, and the knowledge graph.
  • attribute information of the attribute entity determine the recall domain of the information to be searched according to the entity type information; determine the recall content of the information to be searched according to the core information and the attribute information; determine the recall content of the information to be searched according to the recall domain and the recall content to generate the recall condition of the information to be searched; it can be seen that the knowledge map runs through the processing process of query understanding, so that the recall efficiency and recall accuracy can be improved.
  • the query comprehension method for search intent can be understood through multi-granularity (entity recognition granularity (coarse granularity) + word (term) granularity (also called field granularity, ie fine-grained)) structured understanding, entity granularity
  • entity recognition granularity coarse granularity
  • word granularity also called field granularity, ie fine-grained
  • the identification is used to determine the recall domain, and then the fine-grained identification (term granularity) of core entities and/or attribute entities is used to generate recall conditions (or retrieval conditions), which can also improve recall accuracy and recall efficiency.
  • FIG. 1 is a flow chart of an embodiment of a method for understanding a search intent query provided by the present application.
  • Fig. 2 is a schematic structural diagram of an embodiment of an apparatus for understanding a search intent provided by the present application.
  • Fig. 3 is a schematic structural diagram of an embodiment of an electronic device provided by the present application.
  • the search engine set in the network application platform can satisfy the user's query needs.
  • the user can express the active appeal through the query (query) when using the search engine. , and then more targeted recommendation results can be provided through the clearer search intention of the user.
  • the search function is one of the core functions on application service platforms in fields such as e-commerce and local life services.
  • the search link can be divided into the following links: query comprehension, recall, correlation calculation and sorting.
  • the search engine will process and understand the query initiated by the user, including field division, error correction, rewriting, etc. Then recall based on the content understood by the query and calculate the correlation between the query and the doc (that is, the retrieved resources, such as commodities or stores), and finally sort and display the search results to the user.
  • the conventional query understanding only completes the output of the recalled content through the individual functions of each functional module, and there is a certain lack of understanding of the query, and there are also certain defects in the correlation between the recalled content and the actual query content.
  • the present application provides a query comprehension method for search intentions, and the content involved in the embodiments of the present application is used in search engines.
  • a search engine is an online service system that needs to be deployed on a server and use CPU, GPU, etc. to perform multiple calculations.
  • the specific process of the embodiment of the query comprehension method includes the following steps S101 to S104 , and each step will be described in detail below in turn.
  • Step S101 Perform entity identification on the information to be searched, and determine entity type information in the information to be searched.
  • the information to be searched in step S101 may be text information entered in a search box provided on the application service page or information in other forms, such as pictures, videos, voices, and the like.
  • the entity recognition can also be called Named Entity Recognition (Named Entity Recognition), which refers to identifying semantic items with specific meanings in the text, such as: person names, place names, organization names, etc.
  • Named Entity Recognition refers to identifying semantic items with specific meanings in the text, such as: person names, place names, organization names, etc.
  • application service software it can be, for example: dishes, Beverages, medicines, commodities, etc.
  • the semantic item is entity type information, corresponding to different fields in the text information.
  • the search text is "Xiang Zhi Pain Ointment in AA store"
  • the entity recognition result is that AA is the entity type of the pharmacy
  • Xiao Zhong Zhi Pain Ointment is the entity type of medicine, among which: swelling and pain relief are the functional entity types, and ointment is the dosage form entity type
  • the search text is "Kung Pao Chicken located at CC Store on Xueyuan Road”
  • the entity recognition result is that Xueyuan Road is the address entity type
  • CC Store is the store entity type
  • Kung Pao Chicken is the dish entity type.
  • step S101 may include step S101-11 and step S101-12.
  • Step S101-11 Determine whether there is a nested relationship between the entity types of the information to be searched for entity identification.
  • Step S101-12 If there is a nested relationship between the entity types for entity identification of the information to be searched for, determine the entity corresponding to the entity type with the nested relationship as the same type of entity, and determine it as the entity type information .
  • step S101-11 and the step S101-12 are functional entity types, “ointment” is a dosage form entity type, and “Swelling Pain Relief Ointment” is a pharmaceutical entity type as a whole.
  • entity types eg, functional entity types, dosage form entity types
  • entities of the same type eg, drug type
  • Entity recognition will select the largest granularity as the entity recognition result, that is, the final recognition result "Xiaozhongzhipain ointment" is the drug entity type information.
  • the information to be searched also includes the address entity, there is no nested relationship between the address entity type and the drug entity type, so the address entity and the drug entity are two independent entity type information. If the address information is "Beijing Haidian University Road” can identify Beijing Haidian University Road as an address entity without splitting it into multiple entities.
  • the entity identification of the information to be searched for may adopt the maximum granularity (that is, coarse-grained) identification manner.
  • the information to be searched is "a certain pharmacy Loratadine tablet”
  • the entity recognition result is "a certain pharmacy” is the store entity type
  • "loratadine” is the drug name entity type
  • “tablet” is the drug dosage form entity type
  • “Loratadine Tablets” is the drug entity type.
  • entity recognition is a conventional technical means in natural language processing.
  • error correction processing may be performed on the information to be searched, and the fragmentation and / Or complete incomplete information and correct typos.
  • the information to be searched for is "kendej" and can be corrected and processed as "KFC”
  • Amoxicillin can be corrected and processed as "Amoxicillin”, etc.
  • the error correction method is not limited to the above example, any information to be searched can be corrected
  • the operation of completion or adjustment can be error correction, and the purpose is to make the recognition more accurate.
  • error correction can be performed according to the search intention of the information to be searched, and can be combined with the input information and semantics.
  • entity recognition is performed on the error-corrected information to be searched.
  • it may further include: determining the industry type of the information to be searched for.
  • the performing entity identification on the information to be searched and determining the entity type information in the information to be searched includes: performing entity identification on the information to be searched within the scope of the industry type, and determining all the information on the information to be searched The entity type information described above.
  • the industry type of the information to be searched can be determined through the overall text information of the information to be searched, for example: “Swelling and Pain Relief Ointment” can be determined to be related to the pharmaceutical industry, and “Kung Pao Chicken” can be determined to be related to The most important is the catering industry and so on.
  • the scope and difficulty of entity distinction can be reduced during entity recognition. Discriminating the industry before starting to understand the entity knowledge (or before step S102 starts), can also be understood as a classification task at the overall sentence level, and the classification task at the sentence level is more difficult than the sequence labeling task at the word level Low, therefore, it is easier to know the industry scope involved in the information to be searched.
  • Step S102 Acquire the core information of the core entity and the attribute information of the attribute entity according to the association relationship between the core entity and the attribute entity in the entity type information and the knowledge graph.
  • the knowledge graph in step S102 refers to a graph composed of semantically related nodes. For a certain node, the nodes associated with it can be regarded as explaining the knowledge of the node.
  • the knowledge map can be a pre-established data structure, which can be established according to fields, industries, etc.
  • the purpose of the step S102 is to understand the entity type information and obtain entity knowledge.
  • the entity knowledge may include the core information used to describe the core entity in the entity type information and the core information used to describe the entity type information in the entity type information.
  • Attribute information of attribute entities wherein the core entities can be entities specified for different industries or fields, for example: the core entities of the pharmaceutical industry can be drugs, medical devices, pharmacies, etc., and the attribute entities can be applicable diseases, dosage forms, functions etc.; the core entities of the catering industry can be dishes, drinks, restaurants, etc., and the attribute entities can be ingredients, cooking methods, tastes, cuisines, etc.; the core entities of the retail industry can be commodities, supermarkets, and the attribute entities can be brands, materials, etc. ; No more examples here.
  • the entity type information may include core entity type information and/or attribute entity type information, and may also include other entity type information, for example: address entity type information, category entity type information, subject Entity type information, scene entity type information, etc.
  • the core entity type information and attribute entity type information are mainly used for illustration, and entity type information such as address, subject, and category are used as auxiliary descriptions.
  • step S102 includes step S102-1 and step S102-2.
  • Step S102-1 Establish the association relationship between the core entity, the attribute entity, and the entity type corresponding to the knowledge graph.
  • the core entity in the entity type information may include a resource object entity and/or a resource object provider entity
  • an association relationship between the resource object entity and the knowledge graph may be established, and/or, the The association relationship between the resource object provider entity and the knowledge map.
  • Step S102-2 Acquire the core information of the core entity and the attribute information of the attribute entity according to the association relationship.
  • the information of the resource object entity can be obtained according to the association relationship; and/or, the information of the resource object provider entity can be obtained.
  • the resource object entities in the step S102-1 may be service commodities, such as dishes, drinks, medicines, etc.
  • the resource object provider entities may be restaurants, pharmacies, supermarkets, etc.
  • step S102-2 the pharmaceutical industry is used as an example for illustration.
  • “Swelling and Pain Relief Ointment” is linked to the drug corresponding to the knowledge map, and the core entity and attribute entity are understood to obtain entity knowledge, that is, the generic drug name , Applicable diseases, applicable symptoms, drug functions, drug dosage forms, drug ingredients and other information.
  • Step S103 According to the entity type information, determine the recall domain of the information to be searched, and according to the core information and the attribute information to determine the recall content of the information to be searched.
  • Step S104 Generate recall conditions for the information to be searched according to the recall domain and the recall content.
  • the recall content of the information to be searched is determined, specifically
  • the implementation process may include: generating a knowledge type recall condition of the information to be searched according to the recall domain and the recall content.
  • the recall content can be further determined according to the core information of the core entity. Therefore, in the step S103, according to the core information and attribute information, determine the recall content of the information to be searched , may also include steps S103-11 to S103-13.
  • Step S103-11 According to the knowledge map, perform field granularity analysis on the core information and the attribute information as a whole, and obtain unit fields used to describe the core information and the attribute information.
  • the core information "Swelling” and “Analgesic” in the core entity can be obtained as functional components, and the attribute entity "cream” is a dosage form component. Therefore, the three unit fields “relief swelling”, “pain relief” and “ointment” can be obtained by parsing from the field granularity. Among them, detumescence and pain relief can be understood as unit fields of core information, and ointment is a unit field of attribute information. That is to say, the step S103-11 is to analyze the core information and/or the attribute information in a fine-grained manner.
  • Step S103-12 Obtain the rewritten field of the unit field according to the knowledge map.
  • the synonymous words of the unit field in query step S103-1 based on the knowledge map generate synonymous rewriting, and obtain the rewritten field of the unit field, such as "reducing swelling and pain relief paste”, “reducing swelling and pain relief ointment (paste)”, “reducing swelling and pain relief Ointment (paste)” and so on.
  • Step S103-13 Determine the rewritten field as the recall content of the key field in the recall condition.
  • step S104 may include: step S104-11: according to the recall domain and the recall content of the key field, generate a key field type recall condition of the information to be searched.
  • the step S103 of determining recall content of the information to be searched according to the core information and attribute information may further include steps S103-21 to S103-23.
  • Step S103-21 According to the knowledge graph, perform field granularity analysis on the core information and the attribute information as a whole, and obtain unit fields used to describe the core information and the attribute information components.
  • Step S103-22 According to the unit field, determine the weight (term weight) and/or tightness (term tight) of the unit field.
  • Step S103-23 According to the weight (term weight) and/or tightness (term tightness), determine the recall content of the information to be searched.
  • the specific implementation process of the step S103-23 may include: according to the weight and/or compactness of the unit field, Losing and/or rewriting the unit field to obtain a target unit field; determining the target unit field as the recall content.
  • Step S104-21 According to the recall content determined by the recall domain and the weight and/or closeness of the unit field, generate the information to be searched Key field type recall condition. Specifically, the key field type recall condition of the information to be searched may be generated according to the recall domain and the target unit field.
  • the above is the processing procedure when the entity type information includes a core entity and an attribute entity in this embodiment, where the core entity may be a designated entity.
  • the entity type information includes at least one entity of a subject entity, a scene entity, and a category entity
  • the entity type information is tagged, that is, the various character strings searched by the user Normalize to preset categories or tags.
  • tag information corresponding to the at least one type of entity is predicted.
  • the subject entity and scene entity can be determined in conjunction with specific application services, for example: for applications aimed at life services or takeaway services, it can be information related to specific scenes or themes, such as: Mid-Autumn Festival gifts, family gatherings, reunion dinners, etc. ;
  • the category entity can be understood as a category tree with a hierarchical relationship usually maintained for application services.
  • the category tree defines categories according to the tree hierarchy, for example: "Gourmet -> Chinese food -> Local cuisine”. It can also be understood as the category division for a certain service, such as: the division of clothing category "coat, pants, shoes", etc., wherein each category can also include subcategories, such as: shoes can include running shoes, basketball Shoes, casual shoes, leather shoes, etc.
  • step S103 may include: determining the subject entity in the entity type information as a subject recall domain; of hashtags identified as hashtag recall content.
  • step S104 may include: generating a hashtag type recall condition of the information to be searched according to the topic recall field and the hashtag recall content.
  • step S103 may include: determining the scene entity in the entity type information as a scene recall domain; The scene label of is determined as the scene label recall content.
  • step S104 may include: generating the scene label type recall condition of the information to be searched according to the scene recall field and the scene label recall content.
  • step S103 may include: determining the category entity in the entity type information as a category recall domain; The category label of the above category entity is determined as the category label recall content.
  • the specific implementation process of the step S104 may include: recalling the domain and the category according to the category
  • the label recall content generates the category label type recall condition of the information to be searched.
  • step S103 may include: determining the address field in the address entity type information as an address recall field; The standard address name after normalization of the above address fields is determined as the address recall content.
  • step S104 may include: generating the address type recall condition of the information to be searched according to the address recall field and the address recall content.
  • the knowledge type recall condition, the key field type recall condition, the label class recall condition (may include the theme label type recall condition, the scene label type recall condition and/or the category label type recall condition, etc.) and the address Type recall conditions in this embodiment, can be determined as one or more combinations according to the information to be searched, for example: when the entities identified in the information to be searched include entities such as addresses, categories, cores, attributes, etc., then it can be multiple A combination of recall conditions. When the entity identified in the information to be searched includes only one type of entity, the corresponding recall condition can be determined according to the entity.
  • Figure 2 is a schematic structural diagram of an embodiment of a search intent query understanding device provided by the present application.
  • the device embodiment may include a first determining unit 201, a second determining unit 202, and a third determining unit 203.
  • the generating unit 204 may be included in the device embodiment.
  • the first determining unit 201 is configured to perform entity identification on the information to be searched, and determine entity type information in the information to be searched.
  • the first determination unit 201 may specifically include: a nested relationship determination subunit and a determination subunit; the nested relationship determination subunit is configured to determine whether there is nesting between the entity types to be searched for entity identification. nested relationship; the determination subunit is used to determine the entity corresponding to the entity type that exists the nested relationship as the entity of the same type when the determination result of the nested relationship determination subunit is yes, and determine it as the entity type information.
  • the second determination unit 202 is configured to acquire the core information of the core entity and the attribute information of the attribute entity according to the association relationship between the core entity and the attribute entity in the entity type information and the knowledge map.
  • the second determination unit 202 may include: an establishment subunit and an acquisition subunit.
  • the establishing subunit is configured to establish an association relationship between the core entity and attribute entity, and the entity type corresponding to the knowledge graph.
  • the obtaining subunit is configured to obtain the core information of the core entity and the attribute information of the attribute entity according to the association relationship.
  • the third determining unit 203 is configured to determine a recall domain of the information to be searched according to the entity type information; and determine a recall content of the information to be searched according to the core information and the attribute information.
  • the generating unit 204 is configured to generate the recall condition of the information to be searched according to the recall domain and the recall content.
  • the generating unit 204 is specifically configured to generate the knowledge type recall condition of the information to be searched according to the recall domain and the recall content.
  • the third determination unit 203 determines the recall content of the information to be searched according to the core information and the attribute information, which may include: a parsing subunit, Rewrite the subunit, and the content determines the subunit.
  • the parsing subunit is configured to perform field granularity parsing on the core information and the attribute information as a whole according to the knowledge map, and obtain unit fields used to describe the core information and the attribute information.
  • the rewriting subunit is configured to acquire rewritten fields of the unit fields according to the knowledge map.
  • the content determination subunit is configured to determine the rewritten field as the recall content of the key field in the recall condition.
  • the generating unit 204 may specifically generate the key field type recall condition of the information to be searched according to the recall domain and the key field recall content.
  • the third determination unit 203 determines the recall content of the information to be searched according to the core information and attribute information , may include: a parsing subunit, a content determining subunit; wherein, the parsing subunit is configured to perform field granularity parsing on the core information and the attribute information as a whole according to the knowledge graph, and acquire the The unit fields of the core information and the attribute information; the content determination subunit, configured to determine the recall content of the information to be searched according to the weight and/or closeness.
  • the generating unit 204 is configured to generate a key field type recall of the information to be searched according to the recall domain and the recall content determined according to the weight and/or closeness. condition.
  • the content determination subunit includes: an acquisition subunit and a determination subunit; the acquisition subunit is used to perform field loss and/or Or rewritten to obtain the target unit field; the determining subunit is configured to determine the target unit field as the recall content.
  • the generation unit 204 is configured to generate a key field type recall condition of the information to be searched according to the recall domain and the target unit field.
  • it may also include a tagging processing unit and a prediction unit; the tagging processing unit is configured to be used when the entity type information includes at least one entity of a subject entity, a scene entity, and a category entity performing tagging processing on the entity type information; the predicting unit is configured to predict tag information corresponding to the at least one type of entity according to the tagging processing.
  • the third determining unit 203 may specifically determine the subject entity in the entity type information as the subject recall domain; determine the subject tag of the subject entity as the subject tag recall content; the generating unit Specifically, 204 may be to generate a hashtag type recall condition of the information to be searched according to the topic recall field and the hashtag recall content.
  • the third determining unit 203 may specifically determine the scene entity in the entity type information as a scene recall domain; determine the scene label of the scene entity as the scene label recall content; the generating unit Specifically, 204 may be to generate a scene tag type recall condition of the information to be searched according to the scene recall field and the scene tag recall content.
  • the third determining unit 203 may specifically determine the category entity in the entity type information as the category recall domain; determine the category label of the category entity as the category label recall content
  • the generating unit 204 may specifically generate the category label type recall condition of the information to be searched according to the category recall field and the category label recall content.
  • an industry determining unit configured to determine the industry type of the information to be searched; the first determining unit 201 may specifically be used to identify the industry type within the range of the industry type Perform entity identification on the information to be searched, and determine the entity type information in the information to be searched.
  • the industry type of the information to be searched can be determined through the overall text information of the information to be searched, for example: “Swelling and Pain Relief Ointment” can be determined to be related to the pharmaceutical industry, and “Kung Pao Chicken” can be determined to be related to The most important is the catering industry and so on.
  • the scope and difficulty of entity distinction can be reduced during entity recognition. Discriminating the industry before starting to understand the entity knowledge (or before step S102 starts), can also be understood as a classification task at the overall sentence level, and the classification task at the sentence level is more difficult than the sequence labeling task at the word level Low, therefore, it is easier to know the industry scope involved in the information to be searched.
  • an error correction unit configured to perform error correction processing on the information to be searched; the first determination unit 201 may specifically perform an error correction process based on the information to be searched after error correction processing Entity identification, determining the entity type information in the information to be searched.
  • the entity type information when it includes address entity type information, it also includes a fourth determining unit, configured to determine the address field in the address entity type information as an address recall domain; set the address field or The standard address name after the normalization of the address field is determined as the address recall content.
  • the generating unit is configured to generate the address type recall condition of the information to be searched according to the address recall field and the address recall content.
  • the present application also provides a computer storage medium for storing computer program instructions; when the computer program instructions are executed by a processor, execute steps S101 to S104 in the embodiment of the query comprehension method for search intent as described above .
  • the present application also provides an electronic device, and the embodiment of the electronic device includes: a processor 301; a memory 302 for storing computer program instructions, and the computer program instructions are processed When the machine is executed, it executes steps S101 to S104 in the embodiment of the query comprehension method for search intent as described above.
  • a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
  • processors CPUs
  • input/output interfaces network interfaces
  • memory volatile and non-volatile memory
  • Memory may include non-permanent storage in computer readable media, in the form of random access memory (RAM) and/or nonvolatile memory such as read-only memory (ROM) or flash RAM. Memory is an example of computer readable media.
  • RAM random access memory
  • ROM read-only memory
  • flash RAM flash random access memory
  • Computer-readable media include permanent and non-permanent, removable and non-removable media.
  • Information storage can be realized by any method or technology.
  • Information may be computer readable instructions, data structures, modules of a program, or other data.
  • Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Flash memory or other memory technology, Compact Disc Read-Only Memory (CD-ROM), Digital Versatile Disc (DVD) or other optical storage, Magnetic tape cartridge, tape magnetic disk storage or other magnetic storage device or any other non-transmission medium that can be used to store information that can be accessed by a computing device.
  • computer-readable media excludes non-transitory computer-readable media, such as modulated data signals and carrier waves.
  • the embodiments of the present application may be provided as methods, systems or computer program products. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
  • a computer-usable storage media including but not limited to disk storage, CD-ROM, optical storage, etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Animal Behavior & Ethology (AREA)
  • Computational Linguistics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

本申请公开一种搜索意图的查询理解方法和装置、存储介质和电子设备。所述方法包括：对待搜索信息进行实体识别，确定所述待搜索信息中的实体类型信息；根据所述实体类型信息中核心实体和属性实体，与知识图谱之间的关联关系，获取所述核心实体的核心信息和所述属性实体的属性信息；根据所述实体类型信息，确定所述待搜索信息的召回域；根据所述核心信息和所述属性信息，确定所述待搜索信息的召回内容；根据所述召回域和所述召回内容生成所述待搜索信息的召回条件，从而能够提高召回效率以及召回的准确度。

Description

搜索意图的查询理解方法和装置、存储介质和电子设备 技术领域
本申请涉及一种计算机应用技术领域，具体涉及一种搜索意图的查询理解方法和装置。本申请同时涉及一种计算机存储介质和电子设备。
背景技术
为方便用户在网络平台上快速且准确的获取待查找信息，可以通过网络平台设置的搜索引擎输入待查找信息的关键字，进而获取与所述待查找信息相关的信息。
发明内容
本申请提供一种搜索意图的查询理解方法，以解决现有技术中对搜索意图理解不完备，召回效果差的技术问题。
本申请提供一种搜索意图的查询理解方法，包括：对待搜索信息进行实体识别，确定所述待搜索信息中的实体类型信息；根据所述实体类型信息中核心实体和属性实体，与知识图谱之间的关联关系，获取所述核心实体的核心信息和所述属性实体的属性信息；根据所述实体类型信息，确定所述待搜索信息的召回域；根据所述核心信息和所述属性信息，确定所述待搜索信息的召回内容；根据所述召回域和所述召回内容生成所述待搜索信息的召回条件。
在一些实施例中，所述对待搜索信息进行实体识别，确定所述待搜索信息中的实体类型信息，包括：确定所述对待搜索信息进行实体识别的实体类型之间是否存在嵌套关系；若所述对待搜索信息进行实体识别的实体类型之间存在嵌套关系，则将存在所述嵌套关系的实体类型对应的实体作为同一类型实体，确定为所述实体类型信息。
在一些实施例中，所述根据所述实体类型信息中核心实体和属性实体，与知识图谱之间的关联关系，获取所述核心实体的核心信息和所述属性实体的属性信息，包括：建立所述核心实体和属性实体，与所述知识图谱对应的实体类型之间的关联关系；根据所述关联关系，获取所述核心实体的核心信息和所述属性实体的属性信息。
在一些实施例中，所述根据所述召回域和所述召回内容生成所述待搜索信息的召回条件，包括：根据所述召回域和所述召回内容生成所述待搜索信息的知识类型召回条件。
在一些实施例中，所述根据所述核心信息和属性信息，确定所述待搜索信息的召回 内容，包括：根据所述知识图谱，将所述核心信息和所述属性信息作为整体进行字段粒度解析，获取用于描述所述核心信息和所述属性信息的单元字段；根据所述知识图谱，获取所述单元字段的改写字段；将所述改写字段确定为所述召回条件中关键字段召回内容；所述根据所述召回域和所述召回内容生成所述待搜索信息的召回条件，包括：根据所述召回域和所述关键字段召回内容，生成所述待搜索信息的关键字段类型召回条件。
在一些实施例中，所述根据所述核心信息和属性信息，确定所述待搜索信息的召回内容，包括：根据所述知识图谱，将所述核心信息和所述属性信息作为整体进行字段粒度解析，获取用于描述所述核心信息和所述属性信息的单元字段；根据所述单元字段，确定所述单元字段的权重和/或紧密度；所述根据所述权重和/或紧密度，确定所述待搜索信息的召回内容；所述根据所述召回域和所述召回内容生成所述待搜索信息的召回条件，包括：根据所述召回域以及根据所述权重和/或紧密度所确定的召回内容，生成所述待搜索信息的关键字段类型召回条件。
在一些实施例中，所述根据所述权重和/或紧密度，确定所述待搜索信息的召回内容，包括：根据所述单元字段的权重和/或紧密度，对所述单元字段进行丢字段和/或改写，获取目标单元字段；将所述目标单元字段确定为所述召回内容；所述根据所述召回域和所述召回内容生成所述待搜索信息的召回条件，包括：根据所述召回域和所述目标单元字段，生成所述待搜索信息的关键字段类型召回条件。
在一些实施例中，还包括：当所述实体类型信息包括主题实体、场景实体和类目实体中的至少一种实体时，对所述实体类型信息进行标签化处理；根据所述标签化处理，预测与所述至少一种实体对应的标签信息。
在一些实施例中，所述根据所述实体类型信息，确定所述待搜索信息的召回域；根据所述核心信息和所述属性信息，确定所述待搜索信息的召回内容，包括：将所述实体类型信息中的所述主题实体确定为主题召回域；将所述主题实体的主题标签确定为主题标签召回内容；所述根据所述召回域和所述召回内容生成所述待搜索信息的召回条件，包括：根据所述主题召回域和所述主题标签召回内容生成所述待搜索信息的主题标签类型召回条件。
在一些实施例中，所述根据所述实体类型信息，确定所述待搜索信息的召回域；根据所述核心信息和所述属性信息，确定所述待搜索信息的召回内容，包括：将所述实体类型信息中的所述场景实体确定为场景召回域；将所述场景实体的场景标签确定为场景标签召回内容；所述根据所述召回域和所述召回内容生成所述待搜索信息的召回条件，包括：根据所述场景召回域和所述场景标签召回内容生成所述待搜索信息的场景标签类 型召回条件。
在一些实施例中，所述根据所述实体类型信息，确定所述待搜索信息的召回域；根据所述核心信息和所述属性信息，确定所述待搜索信息的召回内容，包括：将所述实体类型信息中的所述类目实体确定为类目召回域；将所述类目实体的类目标签确定为类目标签召回内容；所述根据所述召回域和所述召回内容生成所述待搜索信息的召回条件，包括：根据所述类目召回域和所述类目标签召回内容生成所述待搜索信息的类目标签类型召回条件。
在一些实施例中，还包括：确定所述待搜索信息的行业类型；所述对待搜索信息进行实体识别，确定所述待搜索信息中的实体类型信息，包括：在所述行业类型的范围内对所述待搜索信息进行实体识别，确定所述待搜索信息中的所述实体类型信息。
在一些实施例中，还包括：对所述待搜索信息进行纠错处理；所述对待搜索信息进行实体识别，确定所述待搜索信息中的实体类型信息，包括：根据纠错处理后的所述待搜索信息进行实体识别，确定所述待搜索信息中的所述实体类型信息。
在一些实施例中，还包括：当所述实体类型信息包括地址实体类型信息时，将所述地址实体类型信息中地址字段确定为地址召回域；将所述地址字段或所述地址字段归一化后的标准地址名，确定为地址召回内容；所述根据所述召回域和所述召回内容生成所述待搜索信息的召回条件，包括：根据所述地址召回域和所述地址召回内容生成所述待搜索信息的地址类型召回条件。
本申请还提供一种搜索意图的查询理解装置，包括：第一确定单元，用于对待搜索信息进行实体识别，确定所述待搜索信息中的实体类型信息；第二确定单元，用于根据所述实体类型信息中核心实体和属性实体，与知识图谱之间的关联关系，获取所述核心实体的核心信息和所述属性实体的属性信息；第三确定单元，用于根据所述实体类型信息，确定所述待搜索信息的召回域；根据所述核心信息和所述属性信息，确定所述待搜索信息的召回内容；生成单元，用于根据所述召回域和所述召回内容生成所述待搜索信息的召回条件。
本申请还提供一种计算机存储介质，用于存储计算机程序指令；所述计算机程序指令在被处理器读取执行时，执行如上述搜索意图的查询理解方法的步骤。
本申请还提供一种电子设备，包括：处理器；存储器，用于存储处理器可执行指令的程序，所述程序在被所述处理器读取执行时，执行如上述搜索意图的查询理解方法的步骤。
与现有技术相比，本申请具有以下优点。
本申请提供的一种搜索意图的查询理解方法，通过识别出待搜索信息中的实体类型信息中核心实体和属性实体，与知识图谱之间的关联关系，获取所述核心实体的核心信息和所述属性实体的属性信息；根据所述实体类型信息，确定所述待搜索信息的召回域；根据所述核心信息和所述属性信息，确定所述待搜索信息的召回内容；根据所述召回域和所述召回内容生成所述待搜索信息的召回条件；可见，通过知识图谱贯穿于查询理解的处理过程，从而能够提高召回效率以及召回的准确度。
另外，本申请提供的搜索意图的查询理解方法，能够通过多粒度(实体识别粒度(粗粒度)+词(term)粒度(也可以称为字段粒度，即细粒度))结构化理解，实体粒度的识别用于确定召回域，然后对核心实体和/或属性实体细粒度的识别(term粒度)用于生成召回条件(或者称为检索条件)，同样也能够提高召回的准确度和召回效率。
附图说明
图1是本申请提供一种搜索意图的查询理解方法的实施例的流程图。
图2是本申请提供的一种搜索意图的查询理解装置的实施例的结构示意图。
图3是本申请提供的一种电子设备实施例的结构示意图。
具体实施方式
在下面的描述中阐述了很多具体细节以便于充分理解本申请。但是本申请能够以很多不同于在此描述的其它方式来实施，本领域技术人员可以在不违背本申请内涵的情况下做类似推广，因此本申请不受下面公开的具体实施的限制。
本申请中使用的术语是仅仅出于对特定实施例描述的目的，而非旨在限制本申请。在本申请中和所附权利要求书中所使用的描述方式例如：“一种”、“第一”、和“第二”等，并非对数量上的限定或先后顺序上的限定，而是用来将同一类型的信息彼此区分。
结合上述背景技术可知，通过网络应用平台中设置的搜索引擎满足用户的查询需求，相较于通过推荐系统执行的被动式需求满足，用户在使用搜索引擎时可以通过query(查询)实现主动诉求的表达，进而通过用户较为明确的搜索意图可以提供更具针对性的推荐结果。因为，用户的主动搜索是最直接表达用户真实需求的方式。因此搜索功能在电商、本地生活服务等领域的应用服务平台上属于核心功能之一。
通常搜索链路一般可以划分为如下几个环节：查询理解，召回，相关性计算以及排序。首先搜索引擎会对用户发起的query进行处理和理解，包括分字段、纠错、改写等。 然后基于query理解的内容进行召回并计算query与doc(即，被检索的资源，例如，商品或店铺)的相关性，最后进行排序并把搜索结果展现给用户。然而，常规的query理解仅是通过各个功能模块单独的作用完成召回内容的输出，对query理解存在一定的欠缺，召回内容与实际查询内容之间的相关性也存在一定缺陷。特别是涉及生活服务领域的应用服务平台，查询理解会涉及多种不同的垂直行业，如：餐饮类、生活服务类、医疗服务类、零售类等等，如果按照常规的query理解方式，必然会导致召回内容的相关性较差，召回范围涉及较广导致召回效率低。
鉴于上述，本申请提供一种搜索意图的查询理解方法，本申请的实施例所涉及的内容用于搜索引擎。搜索引擎属于一套在线服务系统，需要部署在服务器上，利用CPU、GPU等进行多项计算。如图1所示，所述查询理解方法实施例的具体过程包括如下步骤S101到步骤S104，下面依次对各个步骤进行详细描述。
步骤S101：对待搜索信息进行实体识别，确定所述待搜索信息中的实体类型信息。
所述步骤S101中所述待搜索信息可以是在应用服务页面中提供的搜索框中输入的文本信息或者其他形式的信息，如图片、视频、语音等。所述实体识别又可以称为命名实体识别(Named Entity Recognition)，指识别文本中有特定含义的语义项，例如：人名，地名，组织机构名等，对于应用服务软件则可以是例如：菜品、饮品、药品、商品等。所述语义项即为实体类型信息，对应文本信息中不同的字段。比如：搜索文本为“AA店铺的消肿止痛膏”，则实体识别结果为AA为药店实体类型，消肿止痛膏为药品实体类型，其中：消肿和止痛是功能实体类型，膏是剂型实体类型；又如：搜索文本为“位于学院路CC店的宫保鸡丁”，则实体识别结果为学院路为地址实体类型，CC店为店铺实体类型，宫保鸡丁为菜品实体类型。
上述内容是对所述步骤S101中涉及的技术名字段的解释以及便于理解的说明示例，并不对技术名字段的适用场景及使用范围进行限定。
所述步骤S101的具体实现过程可以包括步骤S101-11和步骤S101-12。
步骤S101-11：确定所述对待搜索信息进行实体识别的实体类型之间是否存在嵌套关系。
步骤S101-12：若所述对待搜索信息进行实体识别的实体类型之间存在嵌套关系，则将存在所述嵌套关系的实体类型对应的实体作为同一类型实体，确定为所述实体类型信息。
所述步骤S101-11和所述步骤S101-12的目的是对所述待搜索信息进行实体识别时采用最大粒度识别的实体类型信息为准，例如：对待搜索信息为“消肿止痛膏”进行实 体识别。“消肿”和“止痛”是功能实体类型，“膏”是剂型实体类型，“消肿止痛膏”整体是一个药品实体类型。在本申请中，可以合并为同一类型(例如，药品类型)的实体(例如“消肿”、“止痛”和“膏”)对应的实体类型(例如，功能实体类型、剂型实体类型)之间具有嵌套关系。实体识别会选择最大粒度的作为实体识别结果，即最终识别结果“消肿止痛膏”为药品实体类型信息。当然如果待搜索信息中还包括地址实体，而地址实体类型与药品实体类型之间不存在嵌套关系，因此地址实体和药品实体是两个独立的实体类型信息，如果地址信息为“北京海淀学院路”可以将北京海淀学院路识别为一个地址实体，而无需拆分成多个实体。在本实施例中，对于待搜索信息的实体识别可以采用最大粒度(即粗粒度)的识别方式。又例如：待搜索信息为“某某药房氯雷他定片”，实体识别结果为“某某药房”为店铺实体类型，“氯雷他定”为药品名称实体类型，“片”为药品剂型实体类型，而“氯雷他定片”为药品实体类型。
关于对待搜索信息进行实体识别的具体方式此处不再赘述，命名实体识别属于自然语言处理中的常规技术手段。
此处需要说明的是，为了提高实体识别的准确性，在对所述待搜索信息进行命名实体识别时，可以对所述待搜索信息进行纠错处理，将所述待搜索信息中的破碎和/或不完整信息进行补全，对错别字进行修改。例如：待搜索信息为“kendej”可以纠正处理为“肯德基”，“阿莫西”可以纠正处理为“阿莫西林”等，纠错的方式不限于上述举例，任何将所述待搜索信息进行补全或调整的操作均可为纠错，目的在于使得识别更为准确，当然纠错可以根据所述待搜索信息的搜索意图进行纠错，可以结合输入的信息以及语义等等进行纠错。纠错处理后，对纠错后的待搜索信息进行实体识别。
同样地，为提高实体识别的准确性，还可以包括：确定所述待搜索信息的行业类型。
所述对待搜索信息进行实体识别，确定所述待搜索信息中的实体类型信息，包括：在所述行业类型的范围内对所述待搜索信息进行实体识别，确定所述待搜索信息中的所述实体类型信息。
其中，所述待搜索信息的行业类型可以通过待搜索信息整体文本信息进行确定，例如：“消肿止痛膏”即可确定为涉及的是医药行业，“宫保鸡丁”则可以确定为涉及的是餐饮行业等等。
通过对待搜索信息的所属行业的判别，能够在实体识别时减少实体区分的范围和难度。在对实体知识进行理解开始之前对行业进行判别(或者是步骤S102开始之前)，也可以理解为，是对整体句子级别的分类任务，而句子级别的分类任务相对于词级别的序列标注任务难度低，因此，更容易获知所述待搜索信息所涉及的行业范围。
步骤S102：根据所述实体类型信息中核心实体和属性实体，与知识图谱之间的关联关系，获取所述核心实体的核心信息和所述属性实体的属性信息。
所述步骤S102中的知识图谱指存在语义关联的节点组成的图，对某个节点来说，与其关联的节点均可视为解释该节点的知识。知识图谱可以是预先建立的数据结构体，可以是按照领域、行业等建立的。所述步骤S102的目的在对所述实体类型信息进行理解，获得实体知识，所述实体知识可以包括用于描述所述实体类型信息中核心实体的核心信息和用于描述所述实体类型信息中属性实体的属性信息，其中所述核心实体可以是针对不同行业或不同领域指定的实体，例如：医药行业的核心实体可以是药品、医疗器械、药店等，属性实体可以是适用疾病、剂型、功能等；餐饮行业的核心实体可以是菜品、饮品、餐饮店铺等，属性实体可以是食材、烹饪方法、口味、菜系等；零售行业的核心实体可以是商品、超市，属性实体可以是品牌、材质等；此处不再一一举例。换言之，在本实施例中，所述实体类型信息可以包括核心实体类型信息和/或属性实体类型信息，当然还可以包括其他实体类型信息，例如：地址实体类型信息、类目实体类型信息、主题实体类型信息、场景实体类型信息等等，在本实施例中，主要以核心实体类型信息和属性实体类型信息进行举例说明，地址、主题、类目等实体类型信息作为辅助说明。
所述步骤S102的具体实现过程包括步骤S102-1和步骤S102-2。
步骤S102-1：建立所述核心实体和属性实体，与所述知识图谱对应的实体类型之间的关联关系。
其中，所述实体类型信息中的所述核心实体可以包括资源对象实体和/或资源对象提供方实体时，可以建立所述资源对象实体与所述知识图谱的关联关系，和/或，建立所述资源对象提供方实体与所述知识图谱的关联关系。
步骤S102-2：根据所述关联关系，获取所述核心实体的核心信息和所述属性实体的属性信息。
根据步骤S102-2可以根据所述关联关系，获取所述资源对象实体的信息；和/或，获取所述资源对象提供方实体的信息。
所述步骤S102-1中的资源对象实体可以是服务商品，例如菜品、饮品、药品等，资源对象提供方实体可以是餐饮店铺、药店、超市等。
在所述步骤S102-2以所述医药行业为例进行说明，例如“消肿止痛膏”链接到知识图谱对应的药品，对核心实体和属性实体进行理解，获得实体知识，即得到通用药品名、适用疾病、适用症状、药品功能、药品剂型、药品成分等信息。
步骤S103：根据所述实体类型信息，确定所述待搜索信息的召回域，根据所述核心 信息和所述属性信息，确定所述待搜索信息的召回内容。
步骤S104：根据所述召回域和所述召回内容生成所述待搜索信息的召回条件。
在本实施例中，当所述实体类型信息中针对核心实体和属性实体进行实体理解时，所述步骤S104中的根据所述核心信息和属性信息，确定所述待搜索信息的召回内容，具体实现过程可以包括：根据所述召回域和所述召回内容生成所述待搜索信息的知识类型召回条件。
在其他实施例中，还可以根据所述核心实体的核心信息进一步的确定召回内容，因此，所述步骤S103中的所述根据所述核心信息和属性信息，确定所述待搜索信息的召回内容，还可以包括步骤S103-11至S103-13。
步骤S103-11：根据所述知识图谱，将所述核心信息和所述属性信息作为整体进行字段粒度解析，获取用于描述所述核心信息和所述属性信息的单元字段。
例如：对于“消肿止痛膏”，基于知识图谱可以得到核心实体中的核心信息“消肿”、“止痛”是功能成分，属性实体“膏”是剂型成分。因此，从字段粒度进行解析即可得到三个单元字段“消肿”，“止痛”，“膏”。其中，消肿和止痛可以理解为是核心信息的单元字段，膏是属性信息的单元字段。也就是说，所述步骤S103-11是以细粒度方式对核心信息和/或所述属性信息进行解析。
步骤S103-12：根据所述知识图谱，获取所述单元字段的改写字段。
基于知识图谱查询步骤S103-1中的单元字段的同义词产生同义改写，获取单元字段的改写字段，如“消肿止痛贴”，“消肿镇痛膏(贴)”，“消肿去痛膏(贴)”等。
步骤S103-13：将所述改写字段确定为所述召回条件中关键字段召回内容。
所述步骤S104的具体实现可以包括：步骤S104-11：根据所述召回域和所述关键字段召回内容，生成所述待搜索信息的关键字段类型召回条件。
在其他实施例中，所述步骤S103中的所述根据所述核心信息和属性信息，确定所述待搜索信息的召回内容，还可以包括步骤S103-21至步骤S103-23。
步骤S103-21：根据所述知识图谱，将所述核心信息和所述属性信息作为整体进行字段粒度解析，获取用于描述所述核心信息和所述属性信息成分的单元字段。
步骤S103-22：根据所述单元字段，确定所述单元字段的权重(term weight)和/或紧密度(term tight)。
步骤S103-23：根据所述权重(term weight)和/或紧密度(term tight)，确定所述待搜索信息的召回内容。
所述步骤S103-23的具体实现过程可以包括：根据所述单元字段的权重和/或紧密度， 对所述单元字段进行丢字段和/或改写，获取目标单元字段；将所述目标单元字段确定为所述召回内容。
相应的，所述步骤S104的具体实现过程可以包括：步骤S104-21：根据所述召回域和所述单元字段的权重和/或紧密度确定的所述召回内容，生成所述待搜索信息的关键字段类型召回条件。具体地，可以是根据所述召回域和所述目标单元字段，生成所述待搜索信息的关键字段类型召回条件。
以上是对本实施例中，所述实体类型信息中包括核心实体和属性实体时的处理过程，其中，核心实体可以是指定的实体。
那么，当所述实体类型信息中包括主题实体、场景实体和类目实体中的至少一种实体时，对所述实体类型信息进行标签化处理，也就是把用户搜索的多种多样的字符串归一化到预设好的类目或标签上。根据所述标签化处理，预测与所述至少一种实体对应的标签信息。其中，主题实体和场景实体可以是结合具体应用服务确定，例如：对于针对生活服务或者外卖服务的应用，可以是涉及特定场景或主题的信息，如：中秋节礼品、家庭聚会、团圆年夜饭等；而类目实体可以理解为通常对于应用服务维护的具有层级关系的品类树，品类树按照树状层级结构定义类目，例如：“美食->中餐->地方菜”。也可以理解为针对某一服务的类目划分，如：服装类的划分“外套、裤子、鞋”等，其中，针对每个类目还可以包括子类目，如：鞋可以包括跑鞋、篮球鞋、休闲鞋、皮鞋等。
相应地，当所述实体类型信息包括所述主题实体时，所述步骤S103的具体实现过程可以包括：将所述实体类型信息中的所述主题实体确定为主题召回域；将所述主题实体的主题标签确定为主题标签召回内容。
相应地，所述步骤S104的具体实现过程可以包括：根据所述主题召回域和所述主题标签召回内容生成所述待搜索信息的主题标签类型召回条件。
相应地，当所述实体类型信息包括所述场景实体时，所述步骤S103的具体实现过程可以包括：将所述实体类型信息中的所述场景实体确定为场景召回域；将所述场景实体的场景标签确定为场景标签召回内容。
相应地，所述步骤S104的具体实现过程可以包括：根据所述场景召回域和所述场景标签召回内容生成所述待搜索信息的场景标签类型召回条件。
相应地，当所述实体类型信息包括所述类目实体时，所述步骤S103的具体实现过程可以包括：将所述实体类型信息中的所述类目实体确定为类目召回域；将所述类目实体的类目标签确定为类目标签召回内容。
相应地，所述步骤S104的具体实现过程可以包括：根据所述类目召回域和所述类目 标签召回内容生成所述待搜索信息的类目标签类型召回条件。
相应地，当所述实体类型信息包括地址实体类型信息时，所述步骤S103的具体实现过程可以包括：将所述地址实体类型信息中地址字段确定为地址召回域；将所述地址字段或所述地址字段归一化后的标准地址名，确定为地址召回内容。
相应地，所述步骤S104的具体实现过程可以包括：根据所述地址召回域和所述地址召回内容生成所述待搜索信息的地址类型召回条件。
可以理解的是，所述知识类型召回条件、关键字段类型召回条件、标签类召回条件(可以包括主题标签类型召回条件、场景标签类型召回条件和/或类目标签类型召回条件等)以及地址类型召回条件，在本实施例中可以根据待搜索信息确定为一种或多种组合，例如：当待搜索信息中识别的实体包括地址、类目、核心、属性等实体时，则可以是多种召回条件的组合，当待搜索信息中识别的实体仅包括一种实体时，则可以根据所述实体确定对应的召回条件。
以上是对本申请提供的一种搜索意图的查询理解方法实施例的具体描述，与前述提供的一种搜索意图的查询理解方法实施例相对应，本申请还公开一种搜索意图的查询理解装置实施例，请参看图2，由于装置实施例基本相似于方法实施例，所以描述得比较简单，相关之处参见方法实施例的部分说明即可。下述描述的装置实施例仅仅是示意性的。
如图2所示，图2是本申请提供的一种搜索意图的查询理解装置的实施例的结构示意图，装置实施例中可以包括第一确定单元201、第二确定单元202、第三确定单元203、生成单元204。
第一确定单元201，用于对待搜索信息进行实体识别，确定所述待搜索信息中的实体类型信息。
所述第一确定单元201具体可以包括：嵌套关系确定子单元和确定子单元；所述嵌套关系确定子单元，用于确定所述对待搜索信息进行实体识别的实体类型之间是否存在嵌套关系；所述确定子单元，用于在所述嵌套关系确定子单元确定结果为是时，将存在所述嵌套关系的实体类型对应的实体作为同一类型实体，确定为所述实体类型信息。
第二确定单元202，用于根据所述实体类型信息中核心实体和属性实体，与知识图谱的关联关系，获取所述核心实体的核心信息和所述属性实体的属性信息。
所述第二确定单元202可以包括：建立子单元和获取子单元。
所述建立子单元，用于建立所述核心实体和属性实体，与所述知识图谱对应的实体类型之间的关联关系。
所述获取子单元，用于根据所述关联关系，获取所述核心实体的核心信息和所述属性实体的属性信息。
第三确定单元203，用于根据所述实体类型信息，确定所述待搜索信息的召回域；根据所述核心信息和所述属性信息，确定所述待搜索信息的召回内容。
生成单元204，用于根据所述召回域和所述召回内容生成所述待搜索信息的召回条件。
所述生成单元204具体用于根据所述召回域和所述召回内容生成所述待搜索信息的知识类型召回条件。
当所述实体类型信息包括核心实体和属性实体时，所述第三确定单元203中根据所述核心信息和所述属性信息，确定所述待搜索信息的召回内容，可以包括：解析子单元、改写子单元、内容确定子单元。
其中，所述解析子单元，用于根据所述知识图谱，将所述核心信息和所述属性信息作为整体进行字段粒度解析，获取用于描述所述核心信息和所述属性信息的单元字段。
所述改写子单元，用于根据所述知识图谱，获取所述单元字段的改写字段。
所述内容确定子单元，用于将所述改写字段确定为所述召回条件中关键字段召回内容。
相应地，所述生成单元204具体可以是根据所述召回域和所述关键字段召回内容，生成所述待搜索信息的关键字段类型召回条件。
在其他实施例中，为使生成召回条件达到精准的范围，提高查询理解的准确度，所述第三确定单元203中的根据所述核心信息和属性信息，确定所述待搜索信息的召回内容，可以包括：解析子单元、内容确定子单元；其中，所述解析子单元，用于根据所述知识图谱，将所述核心信息和所述属性信息作为整体进行字段粒度解析，获取用于描述所述核心信息和所述属性信息的单元字段；所述内容确定子单元，用于所述根据所述权重和/或紧密度，确定所述待搜索信息的召回内容。
相应地，在该实施例中，所述生成单元204，用于根据所述召回域以及根据所述权重和/或紧密度所确定的召回内容，生成所述待搜索信息的关键字段类型召回条件。
其中，所述内容确定子单元，包括：获取子单元、确定子单元；所述获取子单元，用于根据所述单元字段的权重和/或紧密度，对所述单元字段进行丢字段和/或改写，获取目标单元字段；所述确定子单元，用于将所述目标单元字段确定为所述召回内容。
进一步的，所述生成单元204，用于根据所述召回域和所述目标单元字段，生成所述待搜索信息的关键字段类型召回条件。
在其他实施例中，还可以包括标签化处理单元和预测单元；所述标签化处理单元，用于当所述实体类型信息中包括主题实体、场景实体和类目实体中的至少一种实体时，对所述实体类型信息进行标签化处理；所述预测单元，用于根据所述标签化处理，预测与所述至少一种实体对应的标签信息。
相应地，所述第三确定单元203具体可以是将所述实体类型信息中的所述主题实体确定为主题召回域；将所述主题实体的主题标签确定为主题标签召回内容；所述生成单元204具体可以是根据所述主题召回域和所述主题标签召回内容生成所述待搜索信息的主题标签类型召回条件。
相应地，所述第三确定单元203具体可以是将所述实体类型信息中的所述场景实体确定为场景召回域；将所述场景实体的场景标签确定为场景标签召回内容；所述生成单元204具体可以是根据所述场景召回域和所述场景标签召回内容生成所述待搜索信息的场景标签类型召回条件。
相应地，所述第三确定单元203具体可以是将所述实体类型信息中的所述类目实体确定为类目召回域；将所述类目实体的类目标签确定为类目标签召回内容；所述生成单元204具体可以是根据所述类目召回域和所述类目标签召回内容生成所述待搜索信息的类目标签类型召回条件。
为提高实体识别的准确性，还可以包括：行业确定单元，用于确定所述待搜索信息的行业类型；所述第一确定单元201具体可以用于在所述行业类型的范围内对所述待搜索信息进行实体识别，确定所述待搜索信息中的所述实体类型信息。
其中，所述待搜索信息的行业类型可以通过待搜索信息整体文本信息进行确定，例如：“消肿止痛膏”即可确定为涉及的是医药行业，“宫保鸡丁”则可以确定为涉及的是餐饮行业等等。
通过对待搜索信息的所属行业的判别，能够在实体识别时减少实体区分的范围和难度。在对实体知识进行理解开始之前对行业进行判别(或者是步骤S102开始之前)，也可以理解为，是对整体句子级别的分类任务，而句子级别的分类任务相对于词级别的序列标注任务难度低，因此，更容易获知所述待搜索信息所涉及的行业范围。
为提高实体识别的准确性，还可以包括：纠错单元，用于对所述待搜索信息进行纠错处理；所述第一确定单元201具体可以根据纠错处理后的所述待搜索信息进行实体识别，确定所述待搜索信息中的所述实体类型信息。
在其他实施例中，当所述实体类型信息包括地址实体类型信息时，还包括第四确定单元，用于将所述地址实体类型信息中地址字段确定为地址召回域；将所述地址字段或 所述地址字段归一化后的标准地址名，确定为地址召回内容。所述生成单元，用于根据所述地址召回域和所述地址召回内容生成所述待搜索信息的地址类型召回条件。
以上是对本申请提供的一种搜索意图的查询理解装置实施例的描述，关于装置实施例的具体内容可以参考上述步骤S101到步骤S104的描述，此处不再对重复内容进行描述。
基于上述内容，本申请还提供一种计算机存储介质，用于存储计算机程序指令；所述计算机程序指令在被处理器执行时，执行如上述搜索意图的查询理解方法实施例的步骤S101到步骤S104。
基于上述内容，如图3所示，本申请还提供一种电子设备，该电子设备实施例包括：处理器301；存储器302，用于存储计算机程序指令，所述计算机程序指令在被所述处理器执行时，执行如上述搜索意图的查询理解方法实施例的步骤S101到步骤S104。
在一个典型的配置中，计算设备包括一个或多个处理器(CPU)、输入/输出接口、网络接口和内存。
内存可能包括计算机可读介质中的非永久性存储器，随机存取存储器(RAM)和/或非易失性内存等形式，如只读存储器(ROM)或闪存(flash RAM)。内存是计算机可读介质的示例。
1、计算机可读介质包括永久性和非永久性、可移动和非可移动媒体可以由任何方法或技术来实现信息存储。信息可以是计算机可读指令、数据结构、程序的模块或其他数据。计算机的存储介质的例子包括，但不限于相变内存(PRAM)、静态随机存取存储器(SRAM)、动态随机存取存储器(DRAM)、其他类型的随机存取存储器(RAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、快闪记忆体或其他内存技术、只读光盘只读存储器(CD-ROM)、数字多功能光盘(DVD)或其他光学存储、磁盒式磁带，磁带磁磁盘存储或其他磁性存储设备或任何其他非传输介质，可用于存储可以被计算设备访问的信息。按照本文中的界定，计算机可读介质不包括非暂存电脑可读媒体(transitory media)，如调制的数据信号和载波。
2、本领域技术人员应明白，本申请的实施例可提供为方法、系统或计算机程序产品。因此，本申请可采用完全硬件实施例、完全软件实施例或结合软件和硬件方面的实施例的形式。而且，本申请可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本申请虽然以较佳实施例公开如上，但其并不是用来限定本申请，任何本领域技术 人员在不脱离本申请的精神和范围内，都可以做出可能的变动和修改，因此本申请的保护范围应当以本申请权利要求所界定的范围为准。

Claims (15)

  1. 一种搜索意图的查询理解方法，包括：
    对待搜索信息进行实体识别，确定所述待搜索信息中的实体类型信息；
    根据所述实体类型信息中核心实体和属性实体，与知识图谱之间的关联关系，获取所述核心实体的核心信息和所述属性实体的属性信息；
    根据所述实体类型信息，确定所述待搜索信息的召回域；根据所述核心信息和所述属性信息，确定所述待搜索信息的召回内容；
    根据所述召回域和所述召回内容生成所述待搜索信息的召回条件。
  2. 根据权利要求1所述的搜索意图的查询理解方法，其中，所述对待搜索信息进行实体识别，确定所述待搜索信息中的实体类型信息，包括：
    确定所述对待搜索信息进行实体识别的实体类型之间是否存在嵌套关系；
    若所述对待搜索信息进行实体识别的实体类型之间存在嵌套关系，则将存在所述嵌套关系的实体类型对应的实体作为同一类型实体，确定为所述实体类型信息。
  3. 根据权利要求1所述的搜索意图的查询理解方法，其中，所述根据所述实体类型信息中核心实体和属性实体，与知识图谱之间的关联关系，获取所述核心实体的核心信息和所述属性实体的属性信息，包括：
    建立所述核心实体和属性实体，与所述知识图谱对应的实体类型之间的关联关系；
    根据所述关联关系，获取所述核心实体的核心信息和所述属性实体的属性信息。
  4. 根据权利要求1所述的搜索意图的查询理解方法，其中，所述根据所述召回域和所述召回内容生成所述待搜索信息的召回条件，包括：
    根据所述召回域和所述召回内容生成所述待搜索信息的知识类型召回条件。
  5. 根据权利要求1所述的搜索意图的查询理解方法，其中，所述根据所述核心信息和所述属性信息，确定所述待搜索信息的召回内容，包括：
    根据所述知识图谱，将所述核心信息和所述属性信息作为整体进行字段粒度解析，获取用于描述所述核心信息和所述属性信息的单元字段；
    根据所述知识图谱，获取所述单元字段的改写字段；
    将所述改写字段确定为所述召回条件中关键字段召回内容；
    所述根据所述召回域和所述召回内容生成所述待搜索信息的召回条件，包括：
    根据所述召回域和所述关键字段召回内容，生成所述待搜索信息的关键字段类型召 回条件。
  6. 根据权利要求1所述的搜索意图的查询理解方法，其中，所述根据所述核心信息和所述属性信息，确定所述待搜索信息的召回内容，包括：
    根据所述知识图谱，将所述核心信息和所述属性信息作为整体进行字段粒度解析，获取用于描述所述核心信息和所述属性信息的单元字段；
    根据所述单元字段，确定所述单元字段的权重和/或紧密度；
    所述根据所述权重和/或紧密度，确定所述待搜索信息的召回内容；
    所述根据所述召回域和所述召回内容生成所述待搜索信息的召回条件，包括：
    根据所述召回域以及根据所述权重和/或紧密度所确定的召回内容，生成所述待搜索信息的关键字段类型召回条件。
  7. 根据权利要求6所述的搜索意图的查询理解方法，其中，所述根据所述权重和/或紧密度，确定所述待搜索信息的召回内容，包括：
    根据所述单元字段的权重和/或紧密度，对所述单元字段进行丢字段和/或改写，获取目标单元字段；
    将所述目标单元字段确定为所述召回内容；
    所述根据所述召回域和所述召回内容生成所述待搜索信息的召回条件，包括：
    根据所述召回域和所述目标单元字段，生成所述待搜索信息的关键字段类型召回条件。
  8. 根据权利要求1所述搜索意图的查询理解方法，其中，还包括：
    当所述实体类型信息包括主题实体、场景实体和类目实体中的至少一种实体时，对所述实体类型信息进行标签化处理；
    根据所述标签化处理，预测与所述至少一种实体对应的标签信息。
  9. 根据权利要求8所述的搜索意图的查询理解方法，其中，所述根据所述实体类型信息，确定所述待搜索信息的召回域；根据所述核心信息和所述属性信息，确定所述待搜索信息的召回内容，包括：
    将所述实体类型信息中的所述主题实体确定为主题召回域；将所述主题实体的主题标签确定为主题标签召回内容；
    所述根据所述召回域和所述召回内容生成所述待搜索信息的召回条件，包括：
    根据所述主题召回域和所述主题标签召回内容生成所述待搜索信息的主题标签类型召回条件。
  10. 根据权利要求8或9所述的搜索意图的查询理解方法，其中，所述根据所述实体类型信息，确定所述待搜索信息的召回域；根据所述核心信息和所述属性信息，确定所述待搜索信息的召回内容，包括：
    将所述实体类型信息中的所述场景实体确定为场景召回域；将所述场景实体的场景标签确定为场景标签召回内容；
    所述根据所述召回域和所述召回内容生成所述待搜索信息的召回条件，包括：
    根据所述场景召回域和所述场景标签召回内容生成所述待搜索信息的场景标签类型召回条件。
  11. 根据权利要求10所述的搜索意图的查询理解方法，其中，所述根据所述实体类型信息，确定所述待搜索信息的召回域；根据所述核心信息和所述属性信息，确定所述待搜索信息的召回内容，包括：
    将所述实体类型信息中的所述类目实体确定为类目召回域；将所述类目实体的类目标签确定为类目标签召回内容；
    所述根据所述召回域和所述召回内容生成所述待搜索信息的召回条件，包括：
    根据所述类目召回域和所述类目标签召回内容生成所述待搜索信息的类目标签类型召回条件。
  12. 根据权利要求1所述搜索意图的查询理解方法，其中，还包括：
    确定所述待搜索信息的行业类型；
    所述对待搜索信息进行实体识别，确定所述待搜索信息中的实体类型信息，包括：
    在所述行业类型的范围内对所述待搜索信息进行实体识别，确定所述待搜索信息中的所述实体类型信息。
  13. 根据权利要求1所述搜索意图的查询理解方法，其中，还包括：
    当所述实体类型信息包括地址实体类型信息时，将所述地址实体类型信息中地址字段确定为地址召回域；将所述地址字段或所述地址字段归一化后的标准地址名，确定为地址召回内容；
    所述根据所述召回域和所述召回内容生成所述待搜索信息的召回条件，包括：
    根据所述地址召回域和所述地址召回内容生成所述待搜索信息的地址类型召回条件。
  14. 一种计算机存储介质，其上存储有计算机程序指令；
    所述计算机程序指令在被处理器执行时，执行权利要求1到权利要求13任意一项所述的搜索意图的查询理解方法。
  15. 一种电子设备，包括：
    处理器；
    存储器，其上存储有计算机程序指令，所述计算机程序指令在被所述处理器执行时，执行权利要求1到权利要求13任意一项所述的搜索意图的查询理解方法。
PCT/CN2023/084548 2022-01-29 2023-03-29 搜索意图的查询理解方法和装置、存储介质和电子设备 WO2023143640A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210113219.7A CN114168756B (zh) 2022-01-29 2022-01-29 搜索意图的查询理解方法和装置，存储介质和电子设备
CN202210113219.7 2022-01-29

Publications (3)

Publication Number Publication Date
WO2023143640A1 true WO2023143640A1 (zh) 2023-08-03
WO2023143640A9 WO2023143640A9 (zh) 2023-10-05
WO2023143640A8 WO2023143640A8 (zh) 2023-11-02

Family

ID=80489553

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/084548 WO2023143640A1 (zh) 2022-01-29 2023-03-29 搜索意图的查询理解方法和装置、存储介质和电子设备

Country Status (2)

Country Link
CN (1) CN114168756B (zh)
WO (1) WO2023143640A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114168756B (zh) * 2022-01-29 2022-05-13 浙江口碑网络技术有限公司 搜索意图的查询理解方法和装置，存储介质和电子设备
CN115168436B (zh) * 2022-07-20 2023-08-08 贝壳找房(北京)科技有限公司 查询信息处理方法、电子设备及可读存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130110628A1 (en) * 2011-10-28 2013-05-02 Google Inc. Advertisement determination system and method for clustered search results
CN110795528A (zh) * 2019-09-05 2020-02-14 腾讯科技（深圳）有限公司 一种数据查询方法、装置、电子设备及存储介质
CN111460095A (zh) * 2020-03-17 2020-07-28 北京百度网讯科技有限公司 问答处理方法、装置、电子设备及存储介质
CN113641813A (zh) * 2021-09-01 2021-11-12 上海明略人工智能(集团)有限公司 基于知识图谱的问答系统、方法、电子设备及存储介质
CN114168756A (zh) * 2022-01-29 2022-03-11 浙江口碑网络技术有限公司 搜索意图的查询理解方法和装置，存储介质和电子设备

Family Cites Families (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111881374A (zh) * 2012-12-12 2020-11-03 谷歌有限责任公司 基于组合查询提供搜索结果
CN106649818B (zh) * 2016-12-29 2020-05-15 北京奇虎科技有限公司 应用搜索意图的识别方法、装置、应用搜索方法和服务器
CN109145153B (zh) * 2018-07-02 2021-03-12 北京奇艺世纪科技有限公司 意图类别的识别方法和装置
CN110390054B (zh) * 2019-07-25 2022-11-18 北京百度网讯科技有限公司 兴趣点召回方法、装置、服务器和存储介质
CN110569367A (zh) * 2019-09-10 2019-12-13 苏州大学 一种基于知识图谱的空间关键字查询方法、装置及设备
CN111061859B (zh) * 2019-12-02 2023-09-12 深圳追一科技有限公司 基于知识图谱的数据处理方法、装置和计算机设备
CN113254756B (zh) * 2020-02-12 2024-03-26 百度在线网络技术（北京）有限公司 广告召回方法、装置、设备和存储介质
CN111553162B (zh) * 2020-04-28 2023-09-22 腾讯科技（深圳）有限公司 一种意图识别的方法以及相关装置
CN111625633A (zh) * 2020-05-22 2020-09-04 广东飞企互联科技股份有限公司 基于知识图谱的企业制度问答意图识别方法及装置
CN111708943B (zh) * 2020-06-12 2024-03-01 北京搜狗科技发展有限公司 一种搜索结果展示方法、装置和用于搜索结果展示的装置
CN111800493B (zh) * 2020-06-29 2023-07-28 百度在线网络技术（北京）有限公司 信息内容推送方法、装置、电子设备和存储介质
CN112084405A (zh) * 2020-09-04 2020-12-15 北京字节跳动网络技术有限公司 一种搜索方法、装置及计算机存储介质
CN112685544A (zh) * 2020-12-25 2021-04-20 中国联合网络通信集团有限公司 电信信息的查询方法、装置、设备和介质
CN112434072B (zh) * 2021-01-27 2021-04-30 浙江口碑网络技术有限公司 搜索方法、装置、电子设备及存储介质
CN113255351B (zh) * 2021-06-22 2023-02-03 中国平安财产保险股份有限公司 语句意图识别方法、装置、计算机设备及存储介质
CN113742446A (zh) * 2021-07-16 2021-12-03 华中科技大学 一种基于路径排序的知识图谱问答方法及系统

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130110628A1 (en) * 2011-10-28 2013-05-02 Google Inc. Advertisement determination system and method for clustered search results
CN110795528A (zh) * 2019-09-05 2020-02-14 腾讯科技（深圳）有限公司 一种数据查询方法、装置、电子设备及存储介质
CN111460095A (zh) * 2020-03-17 2020-07-28 北京百度网讯科技有限公司 问答处理方法、装置、电子设备及存储介质
CN113641813A (zh) * 2021-09-01 2021-11-12 上海明略人工智能(集团)有限公司 基于知识图谱的问答系统、方法、电子设备及存储介质
CN114168756A (zh) * 2022-01-29 2022-03-11 浙江口碑网络技术有限公司 搜索意图的查询理解方法和装置，存储介质和电子设备

Also Published As

Publication number Publication date
WO2023143640A8 (zh) 2023-11-02
WO2023143640A9 (zh) 2023-10-05
CN114168756A (zh) 2022-03-11
CN114168756B (zh) 2022-05-13

Similar Documents

Publication Publication Date Title
WO2023143640A1 (zh) 搜索意图的查询理解方法和装置、存储介质和电子设备
US10180967B2 (en) Performing application searches
WO2018177252A1 (zh) 一种基于区块链的数据存储以及查询的方法及装置
JP5679993B2 (ja) クエリを実行する方法およびクエリシステム
US20160357851A1 (en) Natural Language Search With Semantic Mapping And Classification
US20190340199A1 (en) Methods and Systems for Identifying, Selecting, and Presenting Media-Content Items Related to a Common Story
US20140143657A1 (en) Generation of topical subjects from alert search terms
US20130138636A1 (en) Image Searching
WO2022057739A1 (zh) 数据分区存储方法、装置及系统
US9299098B2 (en) Systems for generating a global product taxonomy
CN109002432B (zh) 同义词的挖掘方法及装置、计算机可读介质、电子设备
US11157540B2 (en) Search space reduction for knowledge graph querying and interactions
US12062295B2 (en) Food description processing methods and apparatuses
US11093708B2 (en) Adaptive human to machine interaction using machine learning
US20240061875A1 (en) Identifying content items in response to a text-based request
CN110413608A (zh) 数据查询方法、装置、可读存储介质和程序产品
US11568007B2 (en) Method and apparatus for parsing and representation of digital inquiry related natural language
US8983956B1 (en) Category generalization for search queries
TWI547888B (zh) A method of recording user information and a search method and a server
US9009144B1 (en) Dynamically identifying and removing potential stopwords from a local search query
US11080288B2 (en) Data querying system and method
US20240202798A1 (en) Solving sparse data problems in a recommendation system with freezing start
US20240202797A1 (en) Solving sparse data problems in a recommendation system with cold start
CN117056482A (zh) 基于知识图谱的问答方法、装置和处理器及电子设备
CN116306630A (zh) 业务架构的定位方法、装置、电子设备、介质和程序产品

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23746516

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE