WO2020192534A1 - Procédé de recherche, terminal et support - Google Patents

Procédé de recherche, terminal et support Download PDF

Info

Publication number
WO2020192534A1
WO2020192534A1 PCT/CN2020/080086 CN2020080086W WO2020192534A1 WO 2020192534 A1 WO2020192534 A1 WO 2020192534A1 CN 2020080086 W CN2020080086 W CN 2020080086W WO 2020192534 A1 WO2020192534 A1 WO 2020192534A1
Authority
WO
WIPO (PCT)
Prior art keywords
node
searched
search
generalization
nodes
Prior art date
Application number
PCT/CN2020/080086
Other languages
English (en)
Chinese (zh)
Inventor
陈开济
苏德润
Original Assignee
华为技术有限公司
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 华为技术有限公司 filed Critical 华为技术有限公司
Publication of WO2020192534A1 publication Critical patent/WO2020192534A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/53—Querying
    • G06F16/538—Presentation of query results
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually

Definitions

  • This application relates to the field of search technology, in particular to a search method, terminal and medium.
  • Image search also known as image retrieval, is designed to find specific pictures that users need from the gallery.
  • Traditional image search engines use the surrounding text and title of the webpage where the image is located as the text features of the image, and use text search related technologies to solve the image search problem.
  • the text entered by the user includes the keyword “longan”, while the image tag includes “longan” but not “longan”.
  • the keyword "longan” as the query condition to search in the gallery, you cannot find the picture that the user wants.
  • the present application provides a search method, terminal, and medium to solve the problem that the user's expression form is limited during search, resulting in poor user search experience.
  • this application provides a search method, including:
  • At least one search result is obtained according to the content to be searched, and the search result is different from the key word of the content to be searched;
  • the corresponding reason why the at least one search result is found by the search content is displayed.
  • the user can still obtain at least one search result, thereby reducing the restriction on the user's expression form and improving the user's search experience.
  • the user can also see the corresponding reason for the search result from the content to be searched. Regardless of whether users are satisfied with the search results, they can understand the relevance between the content to be searched and the search results from the corresponding reasons, thereby improving the user's search experience.
  • the search result when the method is applied to image search, the search result includes a picture and a picture tag corresponding to the picture;
  • the search result is different from the keywords of the content to be searched, and specifically includes:
  • the picture tag corresponding to the picture is different from the keyword of the content to be searched.
  • the user can search for at least one picture related to the content to be searched, thereby reducing the restriction on the user’s expression when searching for pictures and improving user search Experience.
  • the search result when the method is applied to a text search, the search result includes text and a text label corresponding to the text;
  • the search result is different from the keywords of the content to be searched, and specifically includes:
  • the text label corresponding to the text is different from the keyword of the content to be searched.
  • the user can search for at least one text related to the content to be searched, thereby reducing the restriction on the user's expression when searching for text and improving the user search experience .
  • the search result when the method is applied to a text search, the search result includes text;
  • the search result is different from the keywords of the content to be searched, and specifically includes:
  • the search string in the text is different from the keywords of the content to be searched, wherein the search string is one of the search words supported by the database to be searched.
  • the user can search for at least one text related to the content to be searched, thereby reducing the form of expression to the user when searching for text Limitations to improve user search experience.
  • the search result is a result obtained by generalization according to the keyword.
  • the keywords of the content to be searched are first generalized into a standard expression form, that is, the search term supported by the database to be searched, and then the standard expression form obtained by generalization is used to search.
  • a standard expression form that is, the search term supported by the database to be searched
  • the standard expression form obtained by generalization is used to search.
  • the keywords of the content to be searched include multiple keywords
  • the search result is a result obtained according to the association relationship of the plurality of keywords, and the association relationship of the plurality of keywords is obtained by generalizing the plurality of keywords.
  • the multiple keywords of the content to be searched are first generalized, and the search is performed according to the relationship between the multiple related words obtained by the generalization.
  • the search is performed according to the relationship between the multiple related words obtained by the generalization.
  • the step of obtaining at least one search result according to the content to be searched includes:
  • the at least one search result is obtained by querying the database to be searched by using the at least one generalized word.
  • the keywords of the content to be searched are first generalized to obtain one or more generalized words. Then use the same generalized words as the search terms supported by the database to be searched for searching. In this way, even if the expression form of the content to be searched by the user changes, even if the user does not know the search terms supported by the database to be searched, at least one search result can be obtained, thereby reducing the restriction on the expression form of the user when searching for text To improve user search experience.
  • the step of generalizing the keyword to obtain at least one generalized word includes:
  • each generalization node is determined as a generalization word.
  • keywords are generalized based on the knowledge graph, which can avoid spending a lot of manpower to construct and maintain a synonym dictionary.
  • Using the knowledge graph to assist the search also decouples the classification capabilities of the knowledge graph and the database to be searched, and improves the scalability of the knowledge graph and the database to be searched.
  • the knowledge graph can better express the semantic relationship between entities, which is conducive to improving search accuracy and user search experience.
  • the generalization node is a label node, and the name of the label node is associated with a preset picture label in the database to be searched or The text labels are the same.
  • the generalization method in this implementation manner may be suitable for use in application scenarios where text tags are used to search for text, or image tags are used to search for pictures.
  • the step of using the at least one input node to find at least one generalization node from the knowledge graph includes:
  • At least one first candidate path is constructed, and each of the first candidate paths includes all Input node, at least one label node, and at least one co-occurrence node; wherein the name of the label node is the same as the preset picture label or text label in the database to be searched, and the label node is the same as at least one of the input node
  • the difference in the number of node levels is within a preset threshold range, and the difference in the number of node levels between the co-occurrence node and all input nodes is within the preset threshold range;
  • the label node on the first candidate path with the shortest semantic distance is determined as a generalized node.
  • the closest correlation between multiple input nodes can be found.
  • the generalized words can be determined according to the association relationship, and then the generalized words are used to search. In this way, it is possible to search for the search results that are most closely related to the keywords of the content to be searched, and improve the search accuracy and user search experience.
  • the step of using the at least one input node to find at least one generalization node from the knowledge graph further includes :
  • each of the second candidate paths includes one input Node, and the label node with the smallest difference in the number of node levels from the input node; wherein the difference in the number of node levels between the co-occurrence node and all input nodes is within a preset threshold range, and the label node
  • the name of is the same as the preset picture label or text label in the database to be searched, and the difference in the number of nodes between the label node and at least one of the input nodes is within a preset threshold range;
  • the label nodes on each of the second candidate paths are respectively determined as generalized nodes corresponding to the input nodes on the second candidate paths.
  • the generalized words corresponding to the input nodes are respectively determined, and then the generalized words are used to search. In this way, it is possible to obtain search results that are closely related to the keywords of the content to be searched, and improve the search accuracy and user search experience.
  • the corresponding search result of the at least one search result from the search content is displayed
  • the reasons include:
  • a generalization reason corresponding to each of the generalization words is displayed, and the generalization reason is generated according to a path from the at least one input node to each generalization node.
  • the generalization reason is generated according to the path from the at least one input node to each generalization node to reflect the association relationship between the keywords of the content to be searched and the generalization words. Show the generalization reason to the user so that the user can understand the relationship between the keywords of the content to be searched and the search results.
  • the step of generating a generalization reason corresponding to each of the generalization words includes:
  • each The node uses each knowledge node on the path from the input node to the generalization node and the relationship between the knowledge nodes to generate a generalization reason corresponding to the generalization word.
  • the generalization reason is generated according to the path from the at least one input node to the corresponding generalization node, so as to reflect the association relationship between the keywords of the content to be searched and the generalization words, so that the user can understand The relationship between the keywords of the content to be searched and the search results.
  • the step of generating a generalization reason corresponding to each of the generalization words includes:
  • the generalization reason is generated according to the path from the at least two input nodes to the corresponding generalization node, so as to reflect the association relationship between the keywords of the content to be searched and the generalization words, thereby enabling the user to Understand the relationship between keywords of the content to be searched and search results.
  • the at least one generalized word is used to query the database to be searched to obtain the search result
  • the steps include:
  • the query obtains at least one first search result, the first search result includes data in the database to be searched that meets a first query condition, and the first query condition is obtained by a combination of the at least one generalized word.
  • each of the generalization words is of one type
  • the at least one first search result is categorized and displayed; wherein the relationship between at least two generalized words in the first query condition is an OR operation.
  • the first search result can be displayed to the user in categories, so that the user can view the search result.
  • the search result is obtained by querying the database to be searched by using the at least one generalized word
  • the steps include:
  • the query obtains at least one second search result, the second search result includes data in the to-be-searched database that meets a second query condition, and the second query condition is obtained by combining the at least one query term.
  • the generalized words obtained by generalization are displayed to the user, and the user can select one or more words according to his or her own ideas, thereby performing a second search. In this way, the search results that users expect can be better searched, and the user's search experience can be improved.
  • the present application provides a terminal including: an input and output module, a memory, and one or more processors; the memory stores one or more computer programs, and the one or more computer programs include instructions, When the instructions are executed by the one or more processors, the terminal is caused to implement any method of the first aspect.
  • the present application provides a computer-readable storage medium, including instructions, which when run on a computer, cause the computer to execute any search method in the first aspect.
  • the user can search for at least one search result, thereby reducing the restriction on the user's expression form and improving the user's search experience.
  • the user can also see the corresponding reason for the search result from the content to be searched on the terminal. Regardless of whether users are satisfied with the search results, they can understand the relevance between the content to be searched and the search results from the corresponding reasons, thereby improving the user's search experience.
  • FIG. 1 is a schematic flowchart of one specific implementation of the search method of this application.
  • step S200 is a schematic flowchart of an implementation manner of step S200 in a specific implementation of the search method of this application;
  • FIG. 3 is a partial schematic diagram of an example of a knowledge graph in the specific implementation of this application.
  • FIG. 4 is a schematic flowchart of an implementation manner of a generalization method based on a knowledge graph in a specific implementation of the search method of this application;
  • FIG. 5 is a schematic diagram of a user interface of a terminal involved in a specific implementation of the search method of this application;
  • FIG. 6 is a schematic diagram of a possible user interface of the terminal when the search method of the present application is applied to an example of text search;
  • FIG. 7 is a partial schematic diagram of another example of a knowledge graph in the specific implementation of this application.
  • FIG. 8 is a schematic diagram of a possible user interface of the terminal when the search method of the present application is applied to an example of image search;
  • FIG. 9 is a schematic flowchart of another implementation manner of the generalization method based on the knowledge graph in a specific implementation of the search method of this application;
  • FIG. 10 is a schematic structural diagram of one of the implementation manners of the terminal of this application.
  • FIG. 11 is a schematic structural diagram of the second implementation manner of the terminal of this application.
  • this application provides a search method.
  • This method can be applied to the terminal.
  • the terminal in this application can be a mobile phone (mobile phone), a tablet computer (Pad), a computer with wireless transceiver function, virtual reality (VR) terminal equipment, augmented reality (Augmented Reality, AR) terminal equipment, wearable Equipment etc.
  • VR virtual reality
  • AR Augmented Reality
  • FIG. 1 is a flowchart of a specific implementation of the search method of this application.
  • the search method may include the following steps S100 to S300.
  • the content to be searched can be in various forms such as text, voice, video, and pictures, and this application does not limit the form of the content to be searched.
  • the content to be searched is text, it can be directly applied to the subsequent processing steps.
  • the content to be searched is in a form other than text, it can be converted into corresponding text, and then applied to subsequent processing steps.
  • the sentence text can be directly used in subsequent processing steps.
  • the automatic speech recognition (ASR) module in the voice assistant can convert the voice input by the user into text, and then apply the converted text to In the subsequent processing steps.
  • the voice assistant in the terminal can usually not only obtain the user's instructions for picture or text search, but also obtain other instructions, such as opening or closing applications, making calls, or sending text messages, etc. . Therefore, in the implementation of acquiring the content to be searched through the voice assistant, the intention recognition module in the voice assistant can also be used to identify the real intention of the user. Specifically, when the voice assistant acquires a piece of speech with unknown intentions, it is first converted into text by the ASR module, and then the intention recognition module uses natural language processing technology to recognize the intention expressed by the text. When it is recognized that the user's intention is a search intention, the voice can be determined as the content to be searched and used in the subsequent processing steps of this application.
  • the subsequent search can be made in the gallery of the terminal.
  • the user's intent is the intent of searching for text
  • it can be inquired in the text library of the terminal or a remote text library in the subsequent.
  • the voice is not the content to be searched, and the subsequent processing steps of this application will not be used for processing.
  • S200 Obtain at least one search result according to the content to be searched, and the search result is different from the keywords of the content to be searched.
  • the converted text can be segmented to obtain the segmented sequence.
  • the word segmentation sequence includes at least one word, and each word is marked with a corresponding part of speech. Then, according to the part of speech of the words, at least one word is selected from the word segmentation sequence and used as a keyword.
  • words whose parts of speech are nouns, verbs, adjectives or adverbs can be filtered out.
  • the search content 1 entered by the user is a text: I want something spicy.
  • V means verb
  • q means quantifier
  • adj means adjective
  • ude means particle. Then, filter out "eat” and "spicy” and determine these two words as keywords for the content to be searched.
  • the database to be searched can be a gallery or a text library.
  • the gallery contains at least one picture, and each picture has one or more picture tags.
  • Picture tags are used to indicate objects, scenes, attributes, or people in the picture.
  • the picture label can be obtained by manual marking or through existing picture recognition technology, and the method for obtaining the picture label is not limited in this application.
  • the text library includes at least one text.
  • each text in the text library can also correspond to one or more text labels. Text tags are used to indicate the main content of the text, the objects involved, events, people, time, location, etc.
  • the text label can be obtained through manual labeling or through existing natural language processing technology, and the method for obtaining the text label is not limited in this application.
  • the picture label corresponding to the picture in the gallery, the text label corresponding to the text in the text library, and the character string contained in the text itself in the text library can all be used as search terms supported by the database to be searched.
  • At least one search result can be obtained by using the above keywords of the content to be searched and the search terms supported by the database to be searched.
  • the database to be searched is a text library.
  • the search results can be obtained by matching the text tags.
  • each search result may include a text and at least one text label corresponding to the text.
  • the aforementioned search results are different from the keywords of the content to be searched, which specifically means that the text tags in the search results are different from the keywords of the content to be searched.
  • each search result includes a text describing Sichuan hot pot, and each text corresponds to the text label "Sichuan hot pot”. It can be seen that the text label "Sichuan hot pot" in the search results is different from the keywords "eat” and "spicy" of the content to be searched 1.
  • the search results can be obtained by matching the character strings contained in the text itself.
  • each search result may include a text, and the text itself includes a search string.
  • the search string is one of the search terms supported by the database to be searched, and it supports string matching.
  • the aforementioned search results are different from the keywords of the content to be searched, which specifically refers to that the search string in the text in the search results is different from the keywords of the content to be searched.
  • each search result includes a piece of text, and each piece of text contains the search string "Sichuan Hot Pot”. It can be seen that the search string "Sichuan hot pot" in the text in the search results is different from the keywords "eat” and "spicy" of the content to be searched 1.
  • the database to be searched is a gallery.
  • each search result may include a picture and at least one picture tag corresponding to the picture.
  • the aforementioned search results are different from the keywords of the content to be searched, which specifically means that the image tags in the search results are different from the keywords of the content to be searched.
  • each search result includes a picture about playing basketball, and each picture corresponds to the picture tag "playing basketball”. It can be seen that the image tag "playing basketball" in the search results is different from the keywords "shoot" and "three pointers" of the content to be searched 1.
  • the aforementioned search result is a result obtained by generalizing keywords of the content to be searched.
  • the keywords of the content to be searched are generalized into a standard expression form, that is, search terms supported by the database to be searched, through generalization.
  • a standard expression form that is, search terms supported by the database to be searched
  • the picture or text that the user wants can be searched.
  • generalizing it can get one or more generalized words, which must be the same as the search terms supported by the database to be searched. Therefore, even if the expression form of the content to be searched by the user changes, even if the user does not know the search terms supported by the database to be searched, by generalizing the keywords of the content to be searched, the keywords and search terms can be relieved.
  • the problem of mismatch between search terms supported by the database improves user search experience.
  • At least one search result step is obtained according to the content to be searched, which may specifically include the following steps S210 to S220.
  • S210 Generalize the keywords to obtain at least one generalized word; wherein each generalized word corresponds to at least one keyword of the content to be searched, and the generalized word corresponds to the keyword different.
  • One keyword can be generalized to get one or more generalized words, and multiple keywords can be generalized to get one or more generalized words. Therefore, for each generalized word, it must correspond to at least one keyword.
  • the keywords of the content to be searched may be the same as the search terms supported by the database to be searched. In this case, the keywords may not be generalized, or the keywords may continue to be generalized. If you do not generalize the keywords, you can directly combine these keywords into query conditions and search in the database to be searched. If you continue to generalize the keywords, you can combine the generalized words and keywords obtained by generalization into query conditions, and then query them in the database to be searched.
  • a synonym dictionary can be used to generalize keywords. Specifically, first, a synonym dictionary is constructed manually.
  • the synonym dictionary includes all search terms supported by the database to be searched. These search terms can be picture tags, text tags, or strings in text.
  • the synonym dictionary also includes the corresponding synonyms for each search term.
  • the keywords extracted from the search content are not directly combined into query conditions, but to check whether these keywords are synonyms of the search terms in the synonym dictionary. If it is a synonym, the keyword is converted into the corresponding standard expression form, that is, a generalized word. Since the generalized words are the same as the search terms supported by the database to be searched, the generalized words are used to form the query conditions, and the images or texts are searched in the database to be searched to obtain search results that satisfy users.
  • the construction of the synonym dictionary is heavy.
  • the construction of thesaurus is done manually.
  • the classification capability of the database to be searched that is, the number of search terms supported in the database to be searched is large
  • the workload of constructing a synonym dictionary is also large.
  • different users and different eras may use different expressions to describe the same thing.
  • a search term may need to manually add many synonyms, which leads to a lot of work to construct a synonym dictionary.
  • the classification ability of the database to be searched is often gradually improved, and its range generally develops from tens or hundreds to tens of thousands.
  • the number of search terms in the database to be searched will gradually increase to tens of thousands from tens to hundreds of when it was initially constructed.
  • different users and different eras may use different expressions to describe the same thing. Therefore, in the stage of maintaining the synonym dictionary, it is also necessary to manually add a large number of synonyms of the original search terms, new search terms and their synonyms. This is one of the reasons for the heavy maintenance workload of the thesaurus.
  • the semantic relationship between different search terms and their synonyms is generally not considered too much.
  • the classification of pictures or texts is more refined, and the semantic scope of the newly added search terms may overlap with the original search terms, which leads to the possibility of generalization of keywords.
  • the picture tag when initially constructing the thesaurus, the picture tag includes "dog".
  • developers will add synonyms for "dog”, such as “house dog”, and specific breeds of dogs, such as “husky” and “Shiba Inu” as synonyms for "dog” and add them to synonyms In the dictionary.
  • synonyms for "dog” such as "house dog”
  • specific breeds of dogs such as “husky” and “Shiba Inu” as synonyms for "dog” and add them to synonyms In the dictionary.
  • synonyms for "husky” such as “Siberian Husky”, “Erha”, etc.
  • the scheme of the thesaurus can't express the complex semantic relationship between entities well, resulting in low accuracy of search results.
  • Different entities may have different semantic relationships, such as the relationship between concept and instance, attribute relationship, etc.
  • these different semantic relations are stored as synonymous relations or unrelated parallel relations, and some semantic relations are not even stored at all, so the semantic relations between entities are lost.
  • the synonyms of the picture tag "dog” include “house dog” and “Shiba Inu”, and “husky” is an independent picture tag, and its synonyms include “Siberian Husky” and “ Erha” and so on.
  • the relationship between “dog” and “husky” actually belongs to the relationship between concept and instance, but at this time, "dog” and “husky” are stored as two independent image tags in the thesaurus, and there is no correlation between the two The juxtaposition.
  • the user wants to search for all dogs they will not be able to search for pictures with the "husky” tag but not the "dog” tag.
  • the picture tags supported by the gallery also include "companion dog". Therefore, in the thesaurus, correspondingly, "companion dog” is regarded as a separate picture tag, and the synonym "pet dog” is added to it.
  • the aforementioned “Shiba Inu” and “Husky” are used as companion dogs.
  • you add "Shiba Inu” and “Husky” as synonyms for the picture tag "Companion Dog” it will obviously be the same as the picture tags "Husky” and "dog”. There is a conflict. For this reason, the usage attribute relationship between "husky” and “companion dog” cannot be stored in the thesaurus. At this time, if the user wants to search for a companion dog, he cannot search for pictures with a "husky” label but not a "companion dog” label.
  • the synonym dictionary cannot well express the complex semantic relationship between entities, and it is difficult to cover all the knowledge involved in the classification of the database to be searched. This leads to low accuracy of search results and omissions, thereby reducing user search experience .
  • the embodiment of the present application also provides another method for generalizing keywords, that is, introducing a knowledge graph to generalize keywords.
  • introducing a knowledge graph to generalize keywords In order to facilitate the understanding of the generalization scheme based on the knowledge graph, the concept of the knowledge graph is briefly introduced below.
  • Knowledge Graph is a structured semantic knowledge base, which aims to describe various concepts, examples, and their relationships in the real world.
  • the knowledge graph can be defined and described by Ontology.
  • a complete ontology framework consists of five parts: concepts, relations, functions, axioms and instances.
  • a lightweight ontology framework can also be used to define and describe a knowledge graph, that is, only a few of the five parts of the ontology framework are used to define and describe a knowledge graph.
  • ontology framework including the aforementioned concepts, relationships, and examples as an example to further illustrate the knowledge graph.
  • Concepts are used to describe actual concepts in general or specialized fields, while instances are basic elements belonging to a concept, representing a concrete object that cannot be subdivided under the current ontology framework. . It should be noted that, depending on the ontology framework, the same thing may be converted between concepts and examples.
  • Relations are used to express the relationship between concepts and concepts, between instances and instances, and between concepts and instances, including the relationship between the upper concept and the lower concept, the relationship between the concept and the instance, the attribute relationship, and the action target relationship.
  • Figure 3 is a partial schematic diagram of a knowledge graph example in an embodiment of this application, which includes 13 knowledge nodes in total: "eat”, “edible”, “food”, “fruit”, “dessert”, “ “Hot pot”, “vegetables”, “Sichuan hot pot”, “Beijing hot pot”, “spicy”, “ma”, “onion” and “tear gas”.
  • “food”, “hot pot”, “spicy”, “ma”, “edible”, “eating”, “fruit”, “dessert”, “vegetable” and “tear gas” are concepts.
  • “Sichuan hot pot”, “Beijing hot pot” and “onion” are examples.
  • the relationship between the concept and the concept can be the relationship between the upper concept and the lower concept, which is represented by "superClassOf” in Figure 3, such as the concept "food” and the concept “hot pot”.
  • the relationship between the concept and the instance can be the relationship between the concept and the instance, which is represented by “hasInstance” in Figure 3, such as the concept “hot pot” and the instance “Sichuan Guohuo”, and the concept “hot pot” and the instance “Beijing Guohuo”.
  • the relationship between the concept and the example can be a taste attribute relationship, which is represented by “hasTaste” in Figure 3, such as the example “Sichuan hot pot” and the concept “spicy”, and the example “Sichuan hot pot” and the concept “ma”.
  • the relationship between the concept and the concept can be the action target relationship, which is represented by “actionTarget” in Figure 3, such as the concept "food” and the concept "food”.
  • the relationship between the concept and the concept can be an alias relationship, which is represented by "aliasOf” in Figure 3, such as the concept "eat” and the concept "edible”.
  • the relationship between the instance and the concept can be a characteristic relationship, which is represented by “hasProperty” in Figure 3, such as the instance "onion” and the concept “tear”.
  • the relationship between the concept and the concept can also be an effect attribute relationship, which is represented by “hasEffect” in Figure 3, for example, the concept "spicy” and the concept "tears.” It should be noted that the specific relationships contained in different knowledge graphs can be preset by the developers who construct the knowledge graphs.
  • the knowledge graph can form a huge semantic network graph, including multiple knowledge nodes (nodes) and multiple edges (edge). Among them, each knowledge node represents an instance or concept, and each edge represents a relationship. If there is a connection path between one knowledge node and another knowledge node, the number of edges between the two knowledge nodes is the number of node layers between the two.
  • concepts and examples can also be collectively referred to as entities, so it can also be understood that each knowledge node represents an entity.
  • the step of generalizing keywords to obtain at least one generalized word based on the knowledge graph may include the following steps S2111 to S2113.
  • S2111 Find at least one input node from the knowledge graph, where the input node is a knowledge node in the knowledge graph, and each input node corresponds to one of the keyword;
  • S2112 Use the at least one input node to find at least one generalized node from the knowledge graph, and the difference in the number of node levels between the generalized node and the input node is within a preset threshold range;
  • the knowledge graph in the embodiment of this application can be obtained in advance from open source knowledge graphs, such as WikiData, Freebase, OpenKG, etc., or it can be pre-built manually according to different application scenarios. This application does not limit the source of the knowledge graph.
  • the process of finding input nodes corresponding to keywords from the knowledge graph is also called entity linking. That is, the keywords of the content to be searched are mapped to the knowledge nodes in the knowledge graph to establish a connection between the keywords and the knowledge graph.
  • Entity connection can include multiple different connection types such as exact matching, alias matching, and prefix and suffix matching.
  • connection types such as exact matching, alias matching, and prefix and suffix matching.
  • priority when performing physical connections, priority can be set for physical connection methods of different connection types. For example, exact matching can be used first, and alias matching can be used when the exact matching cannot be connected to the knowledge node. When the alias matching cannot be connected to the knowledge node, then fuzzy matching methods such as prefix and suffix matching are used.
  • the reason for the entity connection may also be generated, that is, the association relationship between the keyword indicating the content to be searched and the name of the input node.
  • the terminal can show the user the reason for the entity connection, so that the user can more clearly understand the reason for the connection from the keyword of the content to be searched to the corresponding input node in the knowledge graph.
  • a keyword usually at most only one knowledge node corresponding to it can be connected to a knowledge graph, and this knowledge node is called an input node. Sometimes, the keyword may not be connected to any knowledge node. Therefore, every input node found from the knowledge graph must correspond to a keyword.
  • some or all of the knowledge nodes are determined as generalized nodes.
  • all knowledge nodes within the preset threshold range can be determined as generalization nodes. Then the name of each generalization node is determined as a generalization word.
  • the generalized words obtained by generalization in this way are particularly suitable for use in application scenarios where the character string contained in the text is used to search for text.
  • label nodes can be filtered out from all knowledge nodes within a preset threshold range, and all label nodes can be determined as generalized nodes.
  • the label nodes in this application refer to those knowledge nodes in the knowledge graph whose names are the same as the preset picture labels or text labels in the database to be searched. Then the name of each generalization node is determined as a generalization word.
  • the generalized words obtained by generalization in this way are particularly suitable for applications in which text tags are used to search for text or image tags are used to search for pictures.
  • the aforementioned step S2112 may include:
  • At least two input nodes are found from the knowledge graph, and the at least two input nodes have co-occurrence nodes, at least one first candidate path is constructed.
  • the label node on the first candidate path with the shortest semantic distance is determined as a generalized node.
  • a co-occurrence node of multiple input nodes refers to a knowledge node that has a connection path with these multiple input nodes.
  • Each of the above-mentioned first candidate paths includes all input nodes, at least one label node, and at least one co-occurrence node.
  • the difference in the number of node levels between the co-occurrence node and all input nodes is within the preset threshold range.
  • the keywords are examples of "eat” and "spicy".
  • the entity connects it, and a knowledge node “eat” is found from the knowledge graph. The two match, so the knowledge node named "eat” is determined as an input node.
  • the keyword "spicy” it is physically connected, and a knowledge node “spicy” is found from the knowledge graph, and the two match, so the knowledge node named "spicy” is determined as another input node .
  • the preset threshold range is 4 layers. For each input node, find all knowledge nodes within 4 layers of distance, and judge whether the two have co-occurrence nodes.
  • the knowledge nodes within 4 levels of the input node "eating” are: “edible”, “food”, “fruit”, “dessert”, “hot pot”, “vegetable”, “Sichuan hot pot”, “Beijing hot pot” and “onion”.
  • the knowledge nodes within 4 levels of the input node "spicy” are: “Sichuan hot pot”, “ma”, “hot pot”, “Beijing hot pot”, "food”, “fruit”, “dessert”, "edible", “Tear gas”, "onion", "vegetable”.
  • the co-occurrence nodes of the two input nodes are: “Sichuan hot pot”, “hot pot”, “Beijing hot pot”, "food", “fruit”, “dessert”, “edible”, “onion”, and “vegetable”.
  • At least one first candidate path is constructed using two input nodes, co-occurrence nodes of the two input nodes, and label nodes, so that each first candidate path includes all input nodes, at least one label node, and at least A co-occurrence node. Therefore, the first candidate path constructed is as follows:
  • the first candidate path 1 Eating ⁇ Eating ⁇ Food ⁇ Vegetables ⁇ Onion ⁇ Tear ⁇ Spicy.
  • the first candidate path 2 Eat ⁇ Eat ⁇ Food ⁇ Hot Pot ⁇ Sichuan Hot Pot ⁇ Spicy.
  • the semantic distances of the two first candidate paths are calculated respectively.
  • knowledge nodes have corresponding semantic relations, and the semantic distance of the semantic relations has been stored in the knowledge graph in advance.
  • Calculating the semantic distance of the path to be selected mainly refers to the sum of the semantic distances between the knowledge nodes in the path to be selected and calculating its total length.
  • the existing graph traversal algorithm can be used.
  • the shortest path algorithm based on breadth first search can be specifically used.
  • the label node in the first candidate path 2 that is, the label node named "Sichuan Hotpot"
  • the generalized word obtained is "Sichuan hot pot”.
  • the multiple keywords can be generalized to obtain the association relationship of the multiple keywords.
  • the generalization is obtained
  • the generalized word "Sichuan hot pot” in China reflects the correlation between the keywords "eat” and "spicy”. Then use the association relationship of these multiple keywords to search and obtain the corresponding search results. That is, using the generalized word "Sichuan hot pot" to query the database to be searched, since the text label "Sichuan hot pot" originally exists in the database to be searched, at least one search result can be retrieved, and every search result can be Include a text with the text label "Sichuan Hot Pot”.
  • the aforementioned step S2112 may include:
  • the label nodes on each of the second candidate paths are respectively determined as generalized nodes corresponding to the input nodes on the second candidate paths.
  • the co-occurrence node of multiple input nodes refers to a knowledge node that has a connection path with the multiple input nodes.
  • the difference in the number of node levels between the co-occurrence node and all input nodes is within the preset threshold range.
  • each second candidate path may include an input node, and a label node whose difference in the number of nodes from the input node is within a preset threshold range.
  • each second candidate path may include one or more label nodes, or may not include any label nodes. That is, each input node may correspond to zero, one or more generalized nodes. Therefore, after generalization, some keywords do not have corresponding generalized words, and some keywords may correspond to one or more generalized words.
  • the preset threshold range is 2 levels. For each input node, find all knowledge nodes within the range of 2 layers from it, and judge whether the two have co-occurrence nodes.
  • the knowledge nodes within 2 levels of the input node "eat” are: “eat” and “food”.
  • the knowledge nodes within 2 levels from the input node “spicy” are: “Sichuan Hot Pot”, “Ma”, “Hot Pot”, “Tear”, and “Onion”. Therefore, there is no co-occurrence node between the two input nodes.
  • At least one second candidate path is constructed by using two input nodes and a label node, so that each first candidate path includes all input nodes and at least one label node. Since the distance input node "eat” does not have a label node in the knowledge nodes within the second layer, the second candidate path cannot be constructed using the input node "eat”.
  • the distance input node "spicy" includes a label node in the knowledge nodes within the second layer, so at least one second candidate path can be constructed, as follows:
  • the second candidate path 1 onion ⁇ tear ⁇ spicy.
  • the second candidate path 2 Sichuan hot pot ⁇ spicy.
  • the knowledge node "onion” can be determined as a generalized node corresponding to the input node "hot”.
  • the knowledge node "Sichuan hot pot” can also be determined as a generalized node corresponding to the input node "spicy”. Therefore, the generalization method is used to generalize the keywords “eating” and “spicy”, and the generalized words corresponding to the keyword “spicy” are "onion” and "Sichuan hot pot”.
  • each second candidate path may include an input node and a label node with the smallest difference in the number of node levels from the input node.
  • each second candidate path may include one or more label nodes, that is, each input node may correspond to one or more generalized nodes. Therefore, after generalization, there can be one or more generalized words corresponding to each keyword.
  • the example of the knowledge graph in FIG. 3 and the example where the input nodes are "eat” and "spicy” are still used.
  • the preset threshold range is 2 levels.
  • For each input node find all knowledge nodes within the range of 2 layers from it, and judge whether the two have co-occurrence nodes. The result is as before, there is no co-occurrence node between the two.
  • At least one second candidate path is constructed by using two input nodes and a label node, so that each first candidate path includes all input nodes and at least one label node. Since the label node with the smallest difference in the number of nodes from the input node "eat” is "onion", a second candidate path can be constructed using the input node "eat”, as follows:
  • the second candidate path 3 Eating ⁇ Eating ⁇ Food ⁇ Vegetables ⁇ Onion.
  • the second candidate path 4 Sichuan hot pot ⁇ spicy.
  • the knowledge node "onion” can be determined as the generalized node corresponding to the input node "eat”.
  • the knowledge node "Sichuan Hot Pot” can also be determined as a generalized node corresponding to the input node "spicy”. Therefore, using this generalization method to generalize the keywords "eating” and “spicy”, the generalized word corresponding to the keyword “eating” is “onion”, and the generalized word corresponding to the keyword "spicy” The chemical word is "Sichuan hot pot".
  • the scheme of using knowledge graph to generalize keywords has the following advantages.
  • the knowledge graph can adopt automatic or semi-automatic methods to collect and mine knowledge, so as to continuously enrich the existing knowledge graph. Therefore, whether it is in the stage of constructing the knowledge graph or maintaining the knowledge graph, it can be completed in an automated or semi-automated manner. It does not need to be constructed manually for the search terms supported by the database to be searched like the construction and maintenance of a synonym dictionary. And maintenance. In addition, the use of the knowledge graph does not need to consider the overlap of the semantic scope of the newly added search terms with the original search terms, which leads to the problem of conflicts during keyword generalization.
  • the search terms supported by the database to be searched gradually increase accordingly.
  • the synonym dictionary used in generalization is based on the search terms supported by the database to be searched, and the synonym dictionary is always closely related to the classification ability of the database to be searched.
  • the workload and difficulty of maintaining a conflict-free synonym dictionary will become more and more difficult, and the expansion of the system will be very difficult.
  • the knowledge graph and the search terms supported by the database to be searched are independent of each other.
  • the knowledge graph may not be affected, and only the label nodes corresponding to the newly added search terms need to be marked in the knowledge graph.
  • the knowledge graph is continuously enriched, the classification ability of the database to be searched may not be affected. Therefore, the generalization method in this embodiment is adopted to improve the scalability of the system and greatly reduce the expansion cost.
  • the knowledge in the knowledge graph is systematic, and there are many relationships between entities in the knowledge graph, such as the relationship between the upper concept and the lower concept, the relationship between the concept and the instance, the attribute relationship, and the action target relationship. Therefore, the knowledge graph will not lose the semantic relationship between entities like a synonym dictionary, it can better cover the knowledge involved in the classification of the database to be searched, and provide more generalization from the keywords of the content to be searched to generalized words. Good conversion logic.
  • the knowledge node “dog” and the knowledge node “home dog” are alias relationships; the knowledge nodes “dog” and “Shiba Inu” and “huskies” are the relationships between concepts and examples respectively.
  • the knowledge nodes “dog” and “husky” are all tag nodes, that is, the image tags supported by the gallery include the two tags of "dog” and “husky”.
  • the knowledge nodes “dog” and “home dog” are alias relationships; the knowledge nodes “dog” and “Shiba Inu” and “husky” are respectively the relationship between concepts and examples; the knowledge node “companion dog”
  • the relationship with “pet dog” is alias; the relationship between "companion dog” and “Shiba Inu” and “husky” is the use attribute relationship.
  • the knowledge nodes “dog”, “husky” and “companion dog” are all label nodes.
  • the generalization scheme based on the knowledge graph can more accurately search the search results expected by the user, thereby improving the user search experience.
  • the knowledge map better covers the knowledge involved in the classification of the database to be searched, users can more freely use different expressions to search, reducing the restrictions on keywords that users can use, and expanding users’ Form of expression.
  • the aforementioned knowledge graph can also be exported as a synonym dictionary stored locally in the terminal, and the synonym dictionary covers all search terms supported by the database to be searched.
  • the generalization method of the aforementioned thesaurus dictionary is used to generalize the keywords. Using this method can achieve the effect of a generalization scheme based on thesaurus.
  • the generation and maintenance of the thesaurus dictionary is based on the knowledge graph, without human involvement, compared with the original thesaurus-based solution, the solution is more scalable and easy to use.
  • step of S220 Use the at least one generalized word to query the database to be searched to obtain the search result.
  • all generalized words obtained by generalization can be directly combined into the first query condition, and at least one first search result can be obtained by querying in the database to be searched.
  • the first search result includes data that meets the first query condition in the database to be searched.
  • the keywords "eat” and “spicy” are generalized to obtain the generalized word “Sichuan hot pot”.
  • the keywords "eat” and “spicy” are generalized, and the generalized word corresponding to the keyword “eat” is obtained as “onion”, and The generalized word corresponding to the keyword “spicy” is "Sichuan hot pot”.
  • the two generalized words can be combined into the first query condition "Sichuan hot pot or onion", and then search in the text database to obtain at least one text with a text label of "Sichuan hot pot” or "onion”.
  • each generalized word can be defined as one Category, category displays at least one of the aforementioned first search results.
  • this method of classification display can be especially used. For example, when the keyword of the content to be searched is "dog" and there is no "dog” in the search terms supported by the database to be searched, assuming that the two generalized words "Shiba Inu” and "husky” are obtained by generalization, the first query The condition is "Shiba Inu or Husky".
  • the generalized words obtained by the generalization can be displayed to the user, and the user can select one or more words according to their own ideas.
  • the query obtains at least one second search result, and the second search result includes data that meets the second query condition in the database to be searched.
  • the keywords "eat” and “spicy” are generalized, and the generalized word corresponding to the keyword “eat” is obtained as “onion", and the key The generalized word corresponding to the word “spicy” is "Sichuan hot pot”.
  • the two generalized words can be displayed to the user separately for the user to choose. If the user selects "Sichuan hot pot", the query condition "Sichuan hot pot” is directly used to search in the text database, and at least one text with the text label of "Sichuan hot pot” is obtained.
  • the terminal may also display the first search result queried by using all generalized words to the user. If the user is satisfied with the first search result, there is no need to select query terms from generalized terms. If the user wants to filter further, he can select the query words from the generalized words, combine them into the second query condition, and then search to obtain the second search result. At this time, the terminal can display the second search result to the user.
  • the user can also modify the logical relationship between the query terms.
  • the original default relationship between multiple query terms is an OR operation relationship, and the user can modify it to a union operation, or an inversion operation, or a combination of multiple operation relationships, which is not limited in this application.
  • S300 Display the corresponding reason why the at least one search result is searched from the content to be searched.
  • the corresponding reason of at least one search result searched from the content to be searched can reflect the association relationship between the keywords of the content to be searched and the search results.
  • the keywords and generalized words can be formed into a paragraph of text, which is displayed to the user as a corresponding reason, for example, "The synonym for longan is Longan".
  • the generalization path corresponding to the generalized word can be displayed to the user in the form of a picture or text. That is, the generalization reason corresponding to each generalization word is displayed, and the generalization reason is generated according to the path from at least one input node to each generalization node.
  • the construction of the candidate path can refer to the description in the foregoing fourth implementation method of determining generalization nodes. Repeat. For a generalized word, its corresponding generalized node must be in a second candidate path. Therefore, each knowledge node on the path from the input node to the generalization node in the second candidate path and the relationship between these knowledge nodes can be used to generate the generalization reason.
  • the second candidate path 3 eating ⁇ eating ⁇ food ⁇ vegetable ⁇ onion.
  • the path from the input node "eat” to the generalized node “onion” includes a total of 5 knowledge nodes.
  • the relationship between "eating” and “eating” is the alias relationship;
  • the relationship between "eating” and “food” is the relationship of the action goal, and the relationship between "food” and “vegetables” is the upper concept and the lower concept.
  • Relationship: “Vegetables” and “Onions” are also the relationship between upper and lower concepts. Therefore, a piece of text can be generated as the reason for generalization: onion is a kind of vegetable, vegetable is a kind of food, food can be eaten, and eating is an alias for edible.
  • each third candidate path includes an input node. And a label node.
  • each label node on the third path to be selected may be determined as a generalized node, thereby determining at least one generalized word. Similar to the foregoing, for each generalized word, each knowledge node between the input node and the generalized node in the corresponding third candidate path and the relationship between the knowledge nodes can be used to generate the generalization reason.
  • the construction of the candidate path can refer to the description in the third implementation method of determining generalization nodes, which is not here anymore Repeat.
  • its corresponding generalized node must be in a first candidate path. Therefore, for each input node in the first candidate path, each knowledge node on the path from the input node to the generalization node and the relationship between these knowledge nodes can be used to generate a reason fragment. Since the first candidate path contains multiple input nodes, multiple reason segments can be obtained accordingly. By combining these reason fragments, the generalization reason corresponding to the generalization word can be obtained.
  • the first candidate path 2 eating ⁇ eating ⁇ food ⁇ hot pot ⁇ Sichuan hot pot ⁇ spicy.
  • the path from the input node "eat” to the generalization node “Sichuan Hotpot” includes a total of 5 knowledge nodes.
  • the relationship between "eating” and “eating” is the alias relationship
  • the relationship between "eating” and “food” is the relationship of the action goal
  • the relationship between "food” and “hot pot” is the upper concept and the lower concept Relationship
  • the relationship between "hot pot” and “Sichuan hot pot” is also between concepts and examples. Therefore, a piece of text can be generated as a reason fragment: Sichuan hot pot is a kind of hot pot, hot pot is a kind of food, food is edible, and eating is an alias for edible.
  • the path from the input node "spicy” to the generalized node “Sichuan Hotpot” includes a total of 2 knowledge nodes. Among them, the relationship between "eating” and “eating” is the relationship of taste attributes. Therefore, a piece of text can be generated as another reason fragment: Sichuan hot pot tastes spicy.
  • Sichuan hot pot tastes spicy; Sichuan hot pot is a kind of hot pot, hot pot is a kind of food, food is edible, and eating is edible Alias.
  • users can not only see the search results on the terminal, but also see the generalized words and the generalization reasons corresponding to the generalized words. From the reason of generalization, users can understand the input nodes that the keywords of the searched content are connected to in the knowledge graph, as well as the other knowledge nodes experienced from the input node to the generalization node, and the relationship between them , which can help users better obtain the information they want to search, thereby improving user experience.
  • FIG. 5 is a schematic diagram of a user interface of a terminal involved in an embodiment of this application.
  • the user interface is a search interface of the gallery in the terminal.
  • the content to be searched input by the user generalized words obtained by generalization according to the keywords of the content to be searched, and the corresponding generalization reason are displayed in the first area.
  • the first area it is also possible to display the entity connection reason of the keyword of the content to be searched connected to the input node in the knowledge graph.
  • the second area multiple pictures found in the gallery based on generalized words are displayed.
  • the above-mentioned division of the first area and the second area, and the position and size of the first area and the second area in the user interface can all be designed according to actual conditions, which is not limited in this application.
  • FIG. 6 is a schematic diagram of a user interface of a terminal involved in an embodiment of the application.
  • the user interface is a text search interface in the terminal.
  • Use the example of the content to be searched 1 "want to eat something spicy”.
  • There is a search bar in the first area and the search bar is the content to be searched entered by the user "want to eat something spicy".
  • the generalized word that is generalized according to the keywords of the content to be searched namely "Sichuan hot pot" is also displayed.
  • the generalization reason corresponding to the generalization word is also shown, that is, "Sichuan hot pot tastes spicy; Sichuan hot pot is a kind of hot pot, hot pot is a kind of food, food is edible, and eating is an alias for edible ".
  • the second area multiple texts found in the text database based on generalized words are displayed. Each text has at least one text label, and each text has a text label of "Sichuan Hot Pot”.
  • FIG. 7 is a partial schematic diagram of another knowledge graph in an embodiment of this application
  • FIG. 8 is a schematic diagram of a user interface of another terminal involved in an embodiment of this application.
  • the knowledge graph shown in Figure 7 includes a total of 13 knowledge nodes: "throw”, “projection”, “throw”, “action”, “three-point shot”, “basketball rules”, “free throw”, “basketball” "Sports”, “ball”, “sports equipment”, “football”, “basketball” and “basketball”.
  • “Football”, “Basketball” and “Basketball” are label nodes. In this example, except for “soccer”, “basketball” and “playing basketball", the other knowledge nodes are not label nodes.
  • the relationship between the concept and the concept can be the relationship between the upper concept and the lower concept, which is represented by "superClassOf” in Figure 7, such as the concept "ball” and the concept “football”.
  • the relationship between the concept and the instance can be the relationship between the concept and the instance, which is represented by "hasInstance” in Figure 7, such as the concept "basketball rules” and the example "three pointers”.
  • the relationship between the concept and the concept can be the action target relationship, which is represented by "actionTarget” in Figure 7, such as the concept "throwing” and the concept "ball”.
  • the relationship between the concept and the concept can be an alias relationship, which is represented by "aliasOf” in Figure 7, such as the concept “throw” and the concept “throw”.
  • the relationship between the concept and the concept can be a characteristic relationship, which is represented by “hasProperty” in Figure 7, such as the example “basketball” and the concept “basketball rules”.
  • the relationship between the concept and the concept can also be an action domain relationship, which is represented by "actionIn” in Figure 7, such as the concept "throwing” and the concept “basketball”.
  • the relationship between the concept and the instance can be the result relationship, which is represented by "resultIn” in Figure 7, such as the concept "throwing” and the example "three-pointer”.
  • the relationship between the concept and the concept can be a target relationship, which is represented by "targetIn” in Figure 7, such as the concept “basketball” and the concept “basketball sport”. It should be noted that the specific relationships contained in different knowledge graphs can be preset by the developers who construct the knowledge graphs.
  • the database to be searched is a gallery in the terminal, which includes multiple pictures. Each picture has been marked with at least one picture tag through image recognition technology. At least three pictures are labeled as "football”, “basketball” and “playing basketball.”
  • the preset threshold range is 3 levels. For each input node, find all knowledge nodes within a range of 3 layers from it, and judge whether there are co-occurring nodes between the two.
  • the knowledge nodes within the 3 levels of the distance input node "throwing” are: “projection”, “throwing”, “action”, “three pointers”, “basketball rules”, “basketball”, “ball”, “sports” “Equipment”, “Football”, “Basketball” and “Basketball”.
  • the knowledge nodes within the 3 levels of the distance input node "three pointers” are: “throw”, “projection”, “throw”, “action”, “basketball rules”, “free throw”, “basketball”, “ball” “, “sports equipment”, “football”, “basketball” and “playing basketball”. Therefore, the co-occurrence nodes of the two input nodes are: “projection", "pitching”, “action”, “basketball rules”, “basketball”, “ball”, “sports equipment”, “football”, “basketball” And “playing basketball.”
  • At least one first candidate path is constructed using two input nodes, co-occurrence nodes of ten input nodes, and three label nodes, so that each first candidate path includes all input nodes and at least one label node, And at least one co-occurrence node. Therefore, the first candidate path that can be constructed includes at least the following three:
  • the first candidate path 3 Pitch ⁇ Pitch ⁇ Ball ⁇ Football ⁇ Ball ⁇ Pitch ⁇ Three-pointer.
  • the first candidate path 4 throw ⁇ throw ⁇ ball ⁇ basketball ⁇ basketball ⁇ throw ⁇ three-pointer.
  • the first candidate path 5 throw ⁇ throw ⁇ basketball ⁇ play basketball ⁇ basketball ⁇ basketball rules ⁇ three-pointer.
  • the semantic distances of all different first candidate paths are calculated respectively. Assuming that the semantic distance of the first candidate path 5 is the shortest calculated, the label node in the first candidate path 5, that is, the label node named "playing basketball” is determined to be a generalized node. Therefore, this generalization method is used to generalize the keywords “shoot” and "three-point shot", and the generalized word obtained is "play basketball".
  • a generalization reason corresponding to the generalization word "playing basketball” is generated.
  • a total of 4 knowledge nodes are included in the path from the input node “cast” to the generalized node “play basketball”.
  • the relationship between "throwing” and “throwing” is an alias relationship;
  • the relationship between "throwing” and “basketball” is the relationship of action domains, and the relationship between "basketball” and “basketball” is an alias relationship. Therefore, a piece of text can be generated as a reason fragment: throwing is an alias for throwing, the result of throwing may be a three-pointer, throwing is an action in basketball, and playing basketball is an alias for basketball.
  • the path from the input node "three-pointer” to the generalized node "playing basketball” includes a total of 4 knowledge nodes.
  • the relationship between "three pointers” and “basketball rules” is the relationship between concepts and examples; the relationship between “basketball rules” and “basketball” is the characteristic relationship; between "basketball” and “basketball” is Alias relationship. Therefore, a piece of text can be generated as another reason fragment: basketball rules include three-pointers, basketball has basketball rules, and basketball is an alias for basketball.
  • the generalized reason obtained finally is: throwing is an alias for throwing, the result of throwing may be a three-pointer, and throwing is an action in basketball.
  • Basketball rules include three-pointers. Basketball has the characteristics of basketball rules. Basketball is an alias for basketball.
  • the user interface is an interface for image search in the terminal.
  • the search bar is the content to be searched by the user input "find photos of yesterday's three-pointers.”
  • a generalized word that is generalized according to the keywords of the content to be searched, namely "play basketball” is also displayed.
  • the generalization reason corresponding to the generalization word is also displayed, that is, "throwing is an alias for pitching.
  • the result of pitching may be a three-point shot. Pitching is an action in basketball.
  • Basketball rules include three-pointers.
  • each picture has at least one picture tag, and each picture has a "playing basketball" Image tag.
  • the picture label may or may not be displayed on the user interface, which is not limited in this application.
  • the above-mentioned process of generalizing the keywords of the content to be searched can be completed locally in the terminal, or can be completed by one or more servers communicating with the terminal. Online knowledge graphs are stored on the server.
  • the terminal can download the offline knowledge spectrum from the server in advance and store it locally in the terminal, so that after obtaining the content to be searched input by the user, the keywords of the content to be searched can be generalized.
  • the terminal can send the keywords of the content to be searched to the server, and the server uses the online knowledge graph to generalize the keywords, and then generalize the generalized words obtained by the generalization Return to the terminal.
  • the terminal can also directly send the content to be searched to the server, and the server will complete the steps of extracting keywords and generalizing.
  • the generalization scheme performed locally on the terminal can reduce the processing delay.
  • the terminal may include a cache module, or the terminal may be in communication with the cache server.
  • the cache module or the cache server uses the cache module or the cache server, the historical content to be searched, the corresponding keywords, generalization words, generalization reasons, and search results can be stored. If the content to be searched received by the current terminal is the same as the content to be searched in history, or the keywords of the content to be searched are the same as the keywords of the content to be searched in history, the previous generalized words, generalization reasons, and search results can be all Directly display to users directly through the terminal, without the need to perform generalization and search again. In this way, calculation efficiency can be improved, and the time delay for feeding back generalized words, generalized reasons, or search results to users can be shortened.
  • FIG. 9 is a schematic flowchart of another implementation manner of the generalization method based on the knowledge graph.
  • the generalization method based on the knowledge graph may include the following steps S401 to S412.
  • S401 The terminal sends the acquired content to be searched to the generalization module
  • the generalization module sends the content to be searched to the cache server
  • the cache server sends the first information to the generalization module; the first information is used to indicate that the historical content to be searched that is the same as the content to be searched currently is not stored in the cache server;
  • the generalization module After receiving the first information, the generalization module sends the keywords of the content to be searched to the entity connection server;
  • the entity connection server physically connects the keywords in the knowledge graph
  • the entity connection server sends the input node connected in the knowledge graph and the reason for the entity connection to the generalization module;
  • the generalization module sends the input node to the computing server
  • the computing server uses the input node to calculate the generalized node
  • the computing server sends the generalization node and the path from the input node to the corresponding generalization node to the generalization module;
  • the generalization module generates generalization words and corresponding generalization reasons
  • the generalization module sends the generalization word and the corresponding generalization reason to the terminal;
  • the generalization module sends the generalization word and the corresponding generalization reason to the cache server, so that the cache server stores it corresponding to the content to be searched.
  • the entity connection server may be an entity connection server constructed based on an open source search engine, such as ElasticSearch.
  • the entity connection server is used to complete the function of finding the input node most similar to the keyword.
  • the computing server may support the aforementioned semantic distance algorithm, such as the semantic shortest path algorithm.
  • the computing server can use the Graph Engine Service (GES) of Huawei's public cloud.
  • the cache server may adopt Distributed Cache Service (DCS), etc.
  • the generalization module is used as the main control module, which can use interfaces, such as the Representational State Transfer fulfilled (RESTful) interface, for mobile phone gallery search applications or cloud Voice assistants provide explainable knowledge generalization capabilities, namely the aforementioned generalization words and generalization reasons.
  • a terminal is provided.
  • the terminal 600 includes: an input and output module 601, a memory 602, and one or more processors 603; the memory 602 stores one or more computer programs, and the one or more computer programs include instructions. When executed by the one or more processors 603, the terminal 600 is enabled to implement part or all of the steps of any search method in the first embodiment.
  • the input and output module 601 is used to receive user input and display content to the user, such as detecting the user's search input and displaying the search result to the user.
  • the input and output module 601 may include a display screen, a microphone, an input keyboard, and the like.
  • the terminal 600 implements a display function through a graphics processing unit (GPU), a display screen, and an application processor.
  • GPU is a microprocessor for image processing, connected to the display screen and the application processor.
  • the GPU is used to perform mathematical and geometric calculations for graphics rendering.
  • the processor 603 may include one or more GPUs that execute program instructions to generate or change display information.
  • the display screen is used to display images, videos, etc.
  • the display screen includes a display panel.
  • the display panel can adopt liquid crystal display (LCD), organic light-emitting diode (OLED), active-matrix organic light-emitting diode or active-matrix organic light-emitting diode (active-matrix organic light-emitting diode).
  • LCD liquid crystal display
  • OLED organic light-emitting diode
  • active-matrix organic light-emitting diode active-matrix organic light-emitting diode
  • AMOLED flexible light-emitting diode
  • FLED flexible light-emitting diode
  • Miniled MicroLed
  • Micro-oLed quantum dot light-emitting diode
  • QLED quantum dot light-emitting diode
  • the terminal 600 may include 1 or N display screens, and N is a positive integer greater than 1.
  • Microphone also known as “microphone” or “microphone” is used to convert sound signals into electrical signals.
  • the user can make a sound by approaching the microphone with his mouth, and input the sound signal into the microphone.
  • the terminal 600 may be provided with at least one microphone.
  • the terminal 600 may be provided with two microphones, which can implement noise reduction functions in addition to collecting sound signals.
  • the terminal 600 may also be equipped with three, four or more microphones to collect sound signals, reduce noise, identify the source of sound, and realize the function of directional recording.
  • the input keyboard can be a mechanical key keyboard or a touch key keyboard.
  • the terminal 600 can receive the input of the keyboard to obtain the content to be searched input by the user.
  • the memory 602 may include volatile memory (volatile memory), such as random access memory (random access memory, RAM); and may also include non-volatile memory (non-volatile memory), such as flash memory (flash memory), Hard disk drive (HDD) or solid-state drive (SSD); the storage 602 may also include a combination of the above types of storage.
  • volatile memory volatile memory
  • non-volatile memory non-volatile memory
  • flash memory flash memory
  • HDD Hard disk drive
  • SSD solid-state drive
  • the memory 602 may store computer executable program code, and the executable program code includes instructions.
  • the processor 603 executes instructions stored in the memory 602, so as to realize the function or data processing of the terminal.
  • the processor 603 executes program instructions stored in the memory 602 to implement part or all of the steps of any search method in the foregoing embodiments.
  • the memory 602 may include a program storage area and a data storage area.
  • the storage program area can store an operating system, at least one application program (such as a recording function, an image playback function, etc.) required by at least one function.
  • the data storage area can store data (such as audio data, etc.) created during the use of the terminal 600.
  • the processor 603 executes the search function of the terminal by running or executing the computer program or module stored in the memory 602 and calling the code or data stored in the memory 602.
  • the processor 603 may include one or more processing units.
  • the processor 603 may include an application processor (AP), a modem processor, a graphics processing unit (GPU), and an image signal processor. (image signal processor, ISP), controller, video codec, digital signal processor (digital signal processor, DSP), baseband processor, and/or neural-network processing unit (NPU), etc.
  • the different processing units may be independent devices or integrated in one or more processors.
  • the processor may further include a hardware chip.
  • the aforementioned hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD) or a combination thereof.
  • ASIC application-specific integrated circuit
  • PLD programmable logic device
  • the aforementioned PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a general array logic (generic array logic, GAL), or any combination thereof.
  • CPLD complex programmable logic device
  • FPGA field-programmable gate array
  • GAL general array logic
  • the terminal 600 can implement audio functions through an audio module, a microphone, and an application processor, such as recording, using a voice assistant to input content to be searched, and so on.
  • the audio module is used to convert digital audio information into analog audio signal output, and also used to convert analog audio input into digital audio signal.
  • the audio module can also be used to encode and decode audio signals.
  • the audio module may be disposed in the processor 603, or part of the functional modules of the audio module may be disposed in the processor 603.
  • the processor 603 can connect various parts of the terminal 600 by using various interfaces and lines.
  • the processor 603 may include one or more interfaces.
  • the interface may include an integrated circuit (inter-integrated circuit, I2C) interface, an integrated circuit built-in audio (inter-integrated circuit sound, I2S) interface, a pulse code modulation (pulse code modulation, PCM) interface, and a universal asynchronous transmitter (universal asynchronous transmitter) interface.
  • I2C integrated circuit
  • I2S integrated circuit built-in audio
  • PCM pulse code modulation
  • UART universal asynchronous transmitter
  • receiver/transmitter, UART mobile industry processor interface
  • MIPI mobile industry processor interface
  • GPIO general-purpose input/output
  • SIM subscriber identity module
  • USB Universal Serial Bus
  • the structure illustrated in this embodiment does not constitute a specific limitation on the terminal 600.
  • the terminal 600 may include more or less components than the foregoing, or combine certain components, or split certain components, or arrange different components.
  • the aforementioned components can be implemented in hardware, software or a combination of software and hardware.
  • a terminal implementing any search method in the first embodiment is also provided.
  • the terminal can be divided into functional modules.
  • each function module may be divided corresponding to each function, or two or more functions may be integrated into one processing module.
  • the above-mentioned integrated modules can be implemented in the form of hardware or software functional modules. It should be noted that the division of modules in this embodiment is illustrative, and is only a logical function division, and there may be other division methods in actual implementation.
  • FIG. 11 shows another possible structural schematic diagram of the terminal involved in the foregoing embodiment.
  • the terminal may include:
  • the input device 701 is used to obtain the content to be searched input by the user;
  • the processing module 702 is configured to obtain at least one search result according to the content to be searched; wherein the search result is different from the keywords of the content to be searched;
  • the display module 703 is configured to display the corresponding reason why the at least one search result is found by the content to be searched.
  • the aforementioned input device 701 can be a touch screen, an input keyboard, a microphone, etc.
  • the processing module 702 can be one or more processors
  • the display module 703 can be a display screen, a touch screen, a holographic projection device, Virtual reality equipment, etc.
  • the search result when the terminal is used to perform a picture search, the search result includes a picture and a picture tag corresponding to the picture; the picture tag corresponding to the picture is different from the keyword of the content to be searched.
  • the search result when the terminal is used to perform a text search, the search result includes text and a text label corresponding to the text; the text label corresponding to the text is different from the keyword of the content to be searched.
  • the search result when the terminal is used to perform a text search, the search result includes text; the search string in the text is different from the keyword of the content to be searched, and the search string is the search string One of the search terms supported by the database.
  • the search result is a result obtained by generalization according to the keyword.
  • the keywords of the content to be searched include multiple keywords; the search result is a result obtained according to the association relationship of the multiple keywords, and the association relationship of the multiple keywords is determined by the multiple keywords.
  • a keyword is generalized.
  • the processing module 702 is configured to: generalize the keyword to obtain at least one generalized word; and use the at least one generalized word to query the database to be searched to obtain the search result; wherein Each of the generalized words corresponds to at least one keyword of the content to be searched, and the generalized words are different from the corresponding keywords.
  • the processing module 702 is further configured to: find at least one input node from the knowledge graph; use the at least one input node to find at least one generalized node from the knowledge graph; and The name of one generalization node is determined as a generalization word; wherein, the input node is a knowledge node in the knowledge graph, and each input node corresponds to one of the keywords; the generalization The difference in the number of node levels between the node and the input node is within a preset threshold range.
  • the generalized node is a label node, and the name of the label node is the same as a preset picture label or text label in the database to be searched.
  • the processing module 702 is further configured to: when at least two input nodes are found from the knowledge graph, and the at least two input nodes have co-occurrence nodes, construct at least one first pending node. Selecting a path; and, determining a label node on a first candidate path with the shortest semantic distance as a generalized node; wherein each of the first candidate paths includes all input nodes, at least one label node, and At least one co-occurring node; the name of the label node is the same as the preset picture label or text label in the database to be searched, and the difference in the number of node levels between the label node and at least one of the input nodes is within a preset threshold range Within, the difference in the number of node levels between the co-occurrence node and all input nodes is within a preset threshold range.
  • the processing module 702 is further configured to: when at least two input nodes are found from the knowledge graph, and the at least two input nodes do not have co-occurrence nodes, construct at least one second waiting node. Selecting a path; and, determining each label node on the second candidate path as a generalized node corresponding to an input node on the second candidate path; wherein, each of the second candidate paths
  • the selected path includes an input node and a label node with the smallest difference in the number of node levels from the input node; the difference in the number of node levels between the co-occurrence node and all input nodes is within a preset threshold range, so
  • the name of the label node is the same as the preset picture label or text label in the database to be searched, and the difference in the number of node levels between the label node and the at least one input node is within a preset threshold range.
  • the display module 703 is further configured to: display a generalization reason corresponding to each generalization word, wherein the generalization reason is based on the path from the at least one input node to each generalization node generate.
  • the processing module 702 is further configured to: find one input node from the knowledge graph, or find at least two input nodes from the knowledge graph, and the at least two input nodes In the absence of co-occurrence nodes, for each generalized node, each knowledge node on the path from the input node to the generalized node and the relationship between the knowledge nodes are used to generate a relationship with the generalized node.
  • Reasons for generalization corresponding to the chemical words.
  • the processing module 702 is further configured to: in the case that at least two input nodes are found from the knowledge graph, and the at least two input nodes have co-occurrence nodes, for each generalized node , Using each knowledge node on the path from each of the at least two input nodes to the generalization node and the relationship between the knowledge nodes to generate at least two reason segments corresponding to the input nodes; and , Combining the at least two reason fragments into a generalization reason corresponding to the generalization word.
  • the processing module 702 is further configured to obtain at least one first search result from the query, wherein the first search result includes data in the database to be searched that meets a first query condition, and the first query condition is determined by The at least one generalized word is combined;
  • the display module 703 is further configured to display the at least one first search result.
  • the display module 703 is further configured to: in a case where the number of the generalized words is greater than one, classify and display the at least one first search result according to each of the generalized words as a category; Wherein, the relationship between the generalized word and the corresponding keyword is the relationship between the upper concept and the lower concept, or the relationship between the concept and the instance, and the at least two generalized words in the query condition are the OR operation Relationship.
  • the display module 703 is further configured to: display the at least one generalized word; and display at least one second search result;
  • the input device 701 is further configured to: obtain at least one query word selected by the user from the at least one generalized word;
  • the processing module 702 is further configured to obtain at least one second search result from the query, wherein the second search result includes data in the database to be searched that meets a second query condition, and the second query condition is determined by the At least one query term is combined.
  • This embodiment also provides a computer-readable storage medium, including instructions, which when run on a computer, cause the computer to execute part or all of the steps of any search method in the first embodiment.
  • the readable storage medium here may be a magnetic disk, an optical disc, a DVD, a USB, a read-only memory (ROM) or a random access memory (RAM), etc.
  • the application does not limit the specific storage medium form.
  • the computer program product includes one or more computer instructions.
  • the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.
  • the computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from a website, computer, server, or data center.
  • the computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or a data center integrated with one or more available media.
  • the usable medium may be a magnetic medium (for example, a floppy disk, a hard disk, and a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid state disk (SSD)).
  • the terminal and the computer-readable storage medium are used to execute part or all of the steps of any method in the first embodiment, and accordingly have the beneficial effects of the foregoing method, which will not be repeated here.
  • the execution order of each step should be determined by its function and internal logic, and the size of each step sequence number does not mean the order of execution, and does not limit the implementation process of the embodiment.
  • the “plurality” in this specification refers to two or more.
  • words such as “first” and “second” are used to distinguish the same or similar items with basically the same function and effect.
  • the words “first”, “second” and the like do not limit the quantity and execution order, and the words “first” and “second” do not limit the difference.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Library & Information Science (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

L'invention concerne un procédé de recherche, un terminal et un support, se rapportant au domaine de la recherche d'informations. Le procédé de recherche consiste à : acquérir un contenu à rechercher entré par un utilisateur (S100) ; sur la base du contenu à rechercher, obtenir au moins un résultat de recherche, le résultat de recherche étant différent d'un mot-clé du contenu à rechercher (S200) ; et afficher une raison correspondante de l'obtention du ou des résultats de recherche à partir du contenu à rechercher (S300). À l'aide du procédé de recherche susmentionné, l'utilisateur peut obtenir au moins un résultat de recherche même si le contenu à rechercher entré par l'utilisateur est différent du résultat de recherche, ce qui permet de réduire la restriction sur la forme d'expression de l'utilisateur et d'améliorer l'expérience de recherche de l'utilisateur. L'utilisateur peut également voir la raison correspondante de l'obtention du résultat de recherche à partir du contenu à rechercher. Indépendamment du fait que l'utilisateur est satisfait du résultat de recherche, il peut comprendre l'association entre le contenu à rechercher et le résultat de recherche à partir de la raison correspondante, ce qui permet d'améliorer l'expérience de recherche de l'utilisateur.
PCT/CN2020/080086 2019-03-26 2020-03-19 Procédé de recherche, terminal et support WO2020192534A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910233755.9 2019-03-26
CN201910233755.9A CN109992689B (zh) 2019-03-26 2019-03-26 搜索方法、终端及介质

Publications (1)

Publication Number Publication Date
WO2020192534A1 true WO2020192534A1 (fr) 2020-10-01

Family

ID=67131486

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/080086 WO2020192534A1 (fr) 2019-03-26 2020-03-19 Procédé de recherche, terminal et support

Country Status (2)

Country Link
CN (1) CN109992689B (fr)
WO (1) WO2020192534A1 (fr)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113961717A (zh) * 2021-10-26 2022-01-21 上海石湾科技有限公司 一种基于知识图谱的搜寻系统

Families Citing this family (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109992689B (zh) * 2019-03-26 2024-01-30 华为技术有限公司 搜索方法、终端及介质
CN110362749A (zh) * 2019-07-15 2019-10-22 北京百度网讯科技有限公司 搜索结果召回方法、装置、服务器和介质
CN110516061A (zh) * 2019-07-24 2019-11-29 视联动力信息技术股份有限公司 一种数据处理方法、装置和计算机可读存储介质
CN112306252A (zh) * 2019-07-30 2021-02-02 北京搜狗科技发展有限公司 一种数据处理方法、装置和用于数据处理的装置
CN110674308A (zh) * 2019-08-23 2020-01-10 上海科技发展有限公司 基于语法模式的科技词表扩充方法、装置、终端、及介质
CN110688492B (zh) * 2019-09-25 2020-12-08 东南大学 一种基于轻量级索引的知识图谱查询方法
CN110909737A (zh) * 2019-11-14 2020-03-24 武汉虹旭信息技术有限责任公司 图片文字识别方法及系统
CN111104536A (zh) * 2019-12-24 2020-05-05 Oppo广东移动通信有限公司 图片搜索方法、装置、终端及存储介质
CN111382374A (zh) * 2020-02-29 2020-07-07 中国平安人寿保险股份有限公司 一种信息展示方法、装置、电子设备和存储介质
CN111309872B (zh) * 2020-03-26 2023-08-08 北京百度网讯科技有限公司 搜索处理方法、装置及设备
CN111782880B (zh) * 2020-07-10 2023-12-15 聚好看科技股份有限公司 语义泛化方法及显示设备
CN112052332A (zh) * 2020-07-30 2020-12-08 中国建设银行股份有限公司 检索方法、装置、电子设备及可读存储介质
CN112307294B (zh) * 2020-11-02 2024-06-25 北京搜狗科技发展有限公司 一种数据处理方法及装置
CN113555061B (zh) * 2021-07-23 2023-03-14 哈尔滨因极科技有限公司 一种无参考基因组的变异检测的数据工作流处理方法
CN116955758A (zh) * 2022-04-13 2023-10-27 华为技术有限公司 搜索方法和电子设备
CN114741550B (zh) * 2022-06-09 2023-02-10 腾讯科技（深圳）有限公司 图像搜索方法、装置、电子设备和计算机可读存储介质
CN115630144B (zh) * 2022-12-21 2023-04-28 中信证券股份有限公司 一种文档搜索方法、装置及相关设备
CN115827829B (zh) * 2023-02-08 2023-05-02 广州极天信息技术股份有限公司 一种基于本体的搜索意图优化方法及系统
CN117743376B (zh) * 2024-02-19 2024-05-03 蓝色火焰科技成都有限公司 一种数字金融服务的大数据挖掘方法、装置及存储介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106909662A (zh) * 2017-02-27 2017-06-30 腾讯科技（上海）有限公司 知识图谱构建方法及装置
US20170270124A1 (en) * 2016-03-17 2017-09-21 Kabushiki Kaisha Toshiba Data Management Device, Data Management System, and Data Management Method
CN107609152A (zh) * 2017-09-22 2018-01-19 百度在线网络技术（北京）有限公司 用于扩展查询式的方法和装置
CN108804633A (zh) * 2018-06-01 2018-11-13 腾讯科技（深圳）有限公司 基于行为语义知识网络的内容推荐方法
CN108874907A (zh) * 2018-05-25 2018-11-23 北京明略软件系统有限公司 一种数据查询方法和装置、计算机可读存储介质
CN109992689A (zh) * 2019-03-26 2019-07-09 华为技术有限公司 搜索方法、终端及介质

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN100595763C (zh) * 2008-02-26 2010-03-24 华中科技大学 基于自然语言的全文检索系统
US9418121B2 (en) * 2013-03-12 2016-08-16 Google Inc. Search results for descriptive search queries

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170270124A1 (en) * 2016-03-17 2017-09-21 Kabushiki Kaisha Toshiba Data Management Device, Data Management System, and Data Management Method
CN106909662A (zh) * 2017-02-27 2017-06-30 腾讯科技（上海）有限公司 知识图谱构建方法及装置
CN107609152A (zh) * 2017-09-22 2018-01-19 百度在线网络技术（北京）有限公司 用于扩展查询式的方法和装置
CN108874907A (zh) * 2018-05-25 2018-11-23 北京明略软件系统有限公司 一种数据查询方法和装置、计算机可读存储介质
CN108804633A (zh) * 2018-06-01 2018-11-13 腾讯科技（深圳）有限公司 基于行为语义知识网络的内容推荐方法
CN109992689A (zh) * 2019-03-26 2019-07-09 华为技术有限公司 搜索方法、终端及介质

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113961717A (zh) * 2021-10-26 2022-01-21 上海石湾科技有限公司 一种基于知识图谱的搜寻系统

Also Published As

Publication number Publication date
CN109992689B (zh) 2024-01-30
CN109992689A (zh) 2019-07-09

Similar Documents

Publication Publication Date Title
WO2020192534A1 (fr) Procédé de recherche, terminal et support
US11593438B2 (en) Generating theme-based folders by clustering digital images in a semantic space
US10599643B2 (en) Template-driven structured query generation
JP2022505237A (ja) コンテンツアイテム推奨をランク付けするための技術
US7788099B2 (en) Method and apparatus for query expansion based on multimodal cross-vocabulary mapping
US8719249B2 (en) Query classification
US10535106B2 (en) Selecting user posts related to trending topics on online social networks
US11741157B2 (en) Propagating multi-term contextual tags to digital content
WO2023065211A1 (fr) Procédé et appareil d'acquisition d'informations
US20220114361A1 (en) Multi-word concept tagging for images using short text decoder
US20130218866A1 (en) Multimodal graph modeling and computation for search processes
US20180268053A1 (en) Electronic document generation using data from disparate sources
US20220358294A1 (en) Creating and Interacting with Data Records having Semantic Vectors and Natural Language Expressions Produced by a Machine-Trained Model
US11573961B2 (en) Delta graph traversing system
US20200065395A1 (en) Efficient leaf invalidation for query execution
US10579630B2 (en) Content creation from extracted content
CN116569164A (zh) 在内容管理系统中对内容进行智能归类的系统和方法
Khan et al. A relational aggregated disjoint multimedia search results approach using semantics
CN114428834B (zh) 检索方法、装置、电子设备及存储介质
Maree Multimedia context interpretation: a semantics-based cooperative indexing approach
KR20160007057A (ko) 의미 태깅된 자연어 질의의 의도에 부합하는 정보 추출 및 제공 방법 및 시스템
EP4275131A1 (fr) Élimination de résultats de recherche de moteur de recherche informatique intelligent
Lang et al. The next-generation search engine: Challenges and key technologies
CN115828915B (zh) 实体消歧方法、装置、电子设备和存储介质
CN112860940B (zh) 基于描述逻辑知识库上有序概念空间的音乐资源检索方法

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: 20779923

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20779923

Country of ref document: EP

Kind code of ref document: A1