Information retrieval method and apparatus, electronic device, and storage medium

Abstract

The present disclosure provides an information retrieval method and device, electronic equipment and storage medium, and particularly relates to the technical field of artificial intelligence such as natural language processing and deep learning. In the specific scheme, a target retrieval word corresponding to a retrieval sentence is determined; a preset knowledge graph is retrieved based on the target retrieval word to obtain a candidate triple associated with the target retrieval word; the weight of each candidate triple is determined according to the matching degree of each candidate triple with the target retrieval word and the knowledge type corresponding to each candidate triple; the target retrieval word is updated according to the weight of each candidate triple; and the target retrieval result is determined based on the updated target retrieval word. Thus, by reasoning and updating the retrieval word based on the knowledge type corresponding to the candidate triple and the matching degree with the target retrieval word, an accurate and reliable target retrieval result can be determined, the cost required for information retrieval is reduced, and the accuracy and reliability of information retrieval are improved.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, specifically to the fields of artificial intelligence technology such as natural language processing and knowledge graphs, and particularly to an information retrieval method, apparatus, electronic device, and storage medium. Background Technology

[0002] With the rapid development of computer technology, fields such as deep learning, natural language processing, and knowledge graphs have also developed rapidly, and knowledge reasoning technology is being applied more and more widely. For example, in the field of retrieval, when there are no search results matching the search query in the database, knowledge reasoning is usually needed to obtain accurate search results.

[0003] In the field of retrieval technologies, the reliability and accuracy of knowledge reasoning mainly rely on structured knowledge bases; however, acquiring structured knowledge bases is complex and costly. Therefore, improving the accuracy and reliability of retrieval has become an urgent problem to be solved. Summary of the Invention

[0004] This disclosure provides an information retrieval method, apparatus, electronic device, and storage medium.

[0005] This disclosure provides an information retrieval method, including:

[0006] Determine the target search terms corresponding to the search query;

[0007] The preset knowledge graph is searched based on the target search term to obtain candidate triples associated with the target search term;

[0008] The weight of each candidate triple is determined based on the matching degree between each candidate triple and the target search term and the knowledge type corresponding to each candidate triple;

[0009] The target search terms are updated according to the weight of each candidate triple;

[0010] Based on the updated target search terms, determine the target search results.

[0011] In another aspect, this disclosure provides an information retrieval device, comprising:

[0012] The first determining module is used to determine the target search terms corresponding to the search statement;

[0013] The acquisition module is used to search a preset knowledge graph based on the target search term in order to obtain candidate triples associated with the target search term;

[0014] The second determining module is used to determine the weight of each candidate triple based on the matching degree between each candidate triple and the target search term and the knowledge type corresponding to each candidate triple;

[0015] An update module is used to update the target search term according to the weight of each candidate triple;

[0016] The third determination module is used to determine the target search results based on the updated target search terms.

[0017] In another aspect of this disclosure, an electronic device is provided, comprising:

[0018] At least one processor; and

[0019] A memory communicatively connected to the at least one processor; wherein,

[0020] The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the information retrieval method described in one aspect of the above embodiment.

[0021] In another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions and a computer program thereon, the computer instructions being used to cause the computer to perform the information retrieval method described in one aspect of the above-described embodiment.

[0022] In another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the information retrieval method described in one of the above embodiments.

[0023] The information retrieval method, apparatus, electronic device, and storage medium disclosed herein can first determine the target search terms corresponding to the search statement, then search a preset knowledge graph based on the target search terms to obtain candidate triples associated with the target search terms, then determine the weight of each candidate triple based on the matching degree between each candidate triple and the target search terms and the knowledge type corresponding to each candidate triple, and then update the target search terms based on the weight of each candidate triple, and determine the target retrieval result based on the updated target search terms. Therefore, in the retrieval process, the target search terms can be searched in the knowledge graph first to obtain associated candidate triples, and then the target search terms can be updated based on the knowledge type corresponding to the candidate triples and the matching degree with the target search terms, thus determining accurate and reliable target retrieval results. This not only reduces the cost of information retrieval but also improves the accuracy and reliability of information retrieval.

[0024] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0025] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0026] Figure 1 A flowchart illustrating an information retrieval method according to an embodiment of this disclosure;

[0027] Figure 1A A schematic diagram of a candidate triplet provided in an embodiment of this disclosure;

[0028] Figure 2 A flowchart illustrating an information retrieval method according to another embodiment of this disclosure;

[0029] Figure 2A A schematic diagram of a preset knowledge graph library provided in an embodiment of this disclosure;

[0030] Figure 2B A schematic diagram of a preset knowledge graph provided in an embodiment of this disclosure;

[0031] Figure 2C A schematic diagram illustrating an information retrieval process according to an embodiment of this disclosure;

[0032] Figure 3 A flowchart illustrating an information retrieval method provided in yet another embodiment of this disclosure;

[0033] Figure 3A A schematic diagram of a preset knowledge graph provided in an embodiment of this disclosure;

[0034] Figure 4 A flowchart illustrating an information retrieval method according to another embodiment of this disclosure;

[0035] Figure 4A A schematic diagram illustrating an information retrieval process according to an embodiment of this disclosure;

[0036] Figure 4B A schematic diagram of an information retrieval framework provided in an embodiment of this disclosure;

[0037] Figure 5 This is a schematic diagram of the structure of an information retrieval device provided in another embodiment of the present disclosure;

[0038] Figure 6 This is a block diagram of an electronic device used to implement the information retrieval method of the embodiments of this disclosure. Detailed Implementation

[0039] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0040] Artificial intelligence (AI) is the study of enabling computers to simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, and planning). It involves both hardware and software technologies. AI hardware technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, and big data processing. AI software technologies mainly include computer vision, speech recognition, natural language processing, as well as machine learning, deep learning, big data processing, and knowledge graph technologies.

[0041] Natural Language Processing (NLP) is the process by which computers process, understand, and utilize human language (such as Chinese and English). It is an interdisciplinary field combining computer science and linguistics, and is often referred to as computational linguistics. Since natural language is the fundamental characteristic that distinguishes humans from other animals, and without language, human thought would be impossible, NLP embodies the highest task and realm of artificial intelligence. In other words, only when computers possess the ability to process natural language can machines be considered to have achieved true intelligence.

[0042] Knowledge graphs are essentially semantic networks, a graph-based data structure composed of nodes and edges. In a knowledge graph, each node represents an entity existing in the real world, and each edge represents the relationship between entities. Simply put, a knowledge graph is a relational network that connects all different kinds of information, providing the ability to analyze problems from the perspective of "relationships."

[0043] The information retrieval method, apparatus, electronic device, and storage medium of this disclosure are described below with reference to the accompanying drawings.

[0044] The information retrieval method of this disclosure can be executed by the information retrieval device provided in this disclosure, which can be configured in an electronic device.

[0045] Figure 1 This is a flowchart illustrating an information retrieval method provided in an embodiment of this disclosure.

[0046] like Figure 1 As shown, this information retrieval method may include the following steps:

[0047] Step 101: Determine the target search terms corresponding to the search statement.

[0048] The target search term can be one or more, and this disclosure does not limit it.

[0049] The search query can be text or speech. Speech queries can be processed first to obtain text queries, etc. This disclosure does not limit this.

[0050] Understandably, after obtaining the search query, it can be processed by word segmentation, entity recognition, and other methods to determine the corresponding target search terms.

[0051] For example, if the search query is "Wei X's year of Super Boy", by processing the search query, the target search terms can be: Wei X, Super Boy.

[0052] It should be noted that the above examples are merely illustrative and should not be construed as limiting the search statements, target search terms, and methods for determining target search terms in the embodiments of this disclosure.

[0053] Step 102: Search the preset knowledge graph based on the target search term to obtain candidate triples associated with the target search term.

[0054] The preset knowledge graph can be a knowledge graph related to the target search term. This disclosure does not limit this.

[0055] Additionally, candidate triples can include [knowledge type; entity; text]. The knowledge type can be various, such as time, place, person, event, etc. This disclosure does not limit this.

[0056] It is understood that the candidate triples associated with the target search term can be one or more, and this disclosure does not limit this.

[0057] For example, given the search query "Who built the Dujiangyan Irrigation System?" and the target search terms "Dujiangyan" and "built", a search of the pre-defined knowledge graph can retrieve candidate triples related to "Dujiangyan" and "built", such as... Figure 1A As shown, by Figure 1A It can be seen that the candidate triplets are: [256 BC; King Zhaoxiang of Qin; Text 1], [256 BC; Yuzui Diversion Dam; Text 1], [256 BC; Feishayan Spillway; Text 1], [256 BC; Li Bing; Text 2], [256 BC; Chengdu; Text 2], [256 BC; Li Bing; Text 2].

[0058] Among them, text 1 can be " King Zhaoxiang of Qin In 256 BC, Li Bing, the governor of Shu Commandery in the Qin State, and his son, drawing on the water management experience of their predecessors, led the local people in constructing the famous Dujiangyan Irrigation System. The overall plan of Dujiangyan was to divide the Minjiang River into two streams, one of which was diverted into the Chengdu Plain. This served both to divert floods and reduce disasters, and to irrigate farmland, turning a potential hazard into a beneficial one. The main project included... Fish Mouth Dividing Embankment , Feisha Weir Spillway "and the water inlet of the bottle mouth." Text 2 can be "256 BC, the governor of Shu County of the Qin State during the Warring States period." Li Bing The Dujiangyan Irrigation System, built by a large number of people, is located in Sichuan. Chengdu The western part of the plain, west of Dujiangyan City Minjiang River It is located 56 kilometers from Chengdu.

[0059] It should be noted that the above examples are merely illustrative and should not be construed as limiting the target search terms, candidate triples, etc., in the embodiments of this disclosure.

[0060] Step 103: Determine the weight of each candidate triple based on the matching degree between each candidate triple and the target search term and the knowledge type corresponding to each candidate triple.

[0061] Understandably, the higher the match between the candidate triple and the target search term, the greater the weight of the candidate triple.

[0062] For example, if the target search terms are: Wei X, Super Boy, the candidate triplet 1 obtained is: [2007; Super Boy; In 2007, he won the third place in the national finals of Super Boy and officially debuted], and the candidate triplet 2 is: [2007; Super Boy; Wei X participated in the Chengdu division of Super Boy, won the runner-up in the Chengdu division and participated in the national breakout competition, in which he won the third place in the national finals].

[0063] Then, based on semantic matching, similarity calculation, etc., the matching degree between candidate triplet 1 and candidate triplet 2 and the target search term can be determined respectively. For example, through semantic matching, the matching degree of candidate triplet 1 can be determined to be 80% and the matching degree of candidate triplet 2 to be 90%. Since the knowledge type of both candidate triplet 1 and candidate triplet 2 is "time type 2007", it can be determined that the knowledge types they correspond to are the same. Therefore, the weights of candidate triplet 1 and 2 can be determined to be related to the matching degree. For example, the weight of candidate triplet 1 can be 0.45 and the weight of candidate triplet 2 can be 0.55.

[0064] It should be noted that the above examples are merely illustrative and should not be construed as limiting the number, content, matching degree, knowledge type, and weighting of each candidate triple in the embodiments of this disclosure.

[0065] Step 104: Update the target search terms according to the weight of each candidate triplet.

[0066] Optionally, the knowledge types and entities contained in candidate triples with weights greater than a threshold can be combined with the original target search terms to form the updated target search terms. Alternatively, candidate triples can be arranged according to their weights, and the knowledge types and entities contained in the first preset number of candidate triples, along with the original target search terms, can be used as the updated target search terms. Alternatively, the knowledge types and entities contained in the candidate triple with the highest weight, along with the original target search terms, can be combined as the updated target search terms. This disclosure does not limit this approach.

[0067] For example, if the weight of candidate triple 1 is 0.8, the weight of candidate triple 2 is 0.05, and the weight of candidate triple 3 is 0.15, then candidate triple 1 has the largest weight. Therefore, the target search terms can be updated based on candidate triple 1. For instance, the knowledge types and entities contained in candidate triple 1 can be used as new target search terms, and combined with the original target search terms to form the updated target search terms, etc. This disclosure does not limit this.

[0068] Alternatively, the different knowledge types contained in the candidate triples can be used as new target search terms. For example, if candidate triple 1 and candidate triple 2 both have a weight of 0.5, and the knowledge type in candidate triple 1 is "Hunan" and the knowledge type in candidate triple 2 is "2007", then if the original target search term is "Super Boy", the updated target search term could be: 2007 Hunan Super Boy.

[0069] It should be noted that the above examples are merely illustrative and should not be construed as limiting the number of total candidate triples, their weights, or the updated target search terms in the embodiments of this disclosure.

[0070] It is understood that in this embodiment of the disclosure, the target search term can be retrieved by searching in a preset knowledge graph to obtain associated candidate triples. Then, the target search term is updated according to the weight of each candidate triple, thereby making the updated target search term more accurate and reliable. That is, in the process of updating the target search term, there is no need to rely on a large-scale knowledge base. By reasoning and updating the search term based on the knowledge type corresponding to the candidate triple and its matching degree with the target search term, the target search term can be updated. This process does not require building a large-scale knowledge base, thereby reducing the complexity and cost of building a large-scale knowledge model base.

[0071] Step 105: Determine the target search results based on the updated target search terms.

[0072] For example, if the knowledge type corresponding to the candidate triple with the highest weight is the time "2007", and the target search term determined based on the search query is "Wei X Super Boy", then the updated target search term could be "2007 Super Boy". Then, "2007 Super Boy" can be used to search within a pre-defined knowledge graph to determine the target search results.

[0073] It should be noted that the above examples are merely illustrative and should not be construed as limiting the updated target search terms, target search results, etc., in the embodiments of this disclosure.

[0074] In this embodiment, the target search term corresponding to the search query is first determined. Then, a preset knowledge graph is searched based on the target search term to obtain candidate triples associated with the target search term. Next, the weight of each candidate triple is determined according to the matching degree between each candidate triple and the target search term, and the knowledge type corresponding to each candidate triple. Then, the target search term is updated based on the weight of each candidate triple. Based on the updated target search term, the target search result is determined. Therefore, in the retrieval process, the target search term can be searched in the knowledge graph first to obtain associated candidate triples. Then, the target search term is updated based on the knowledge type corresponding to the candidate triple and the matching degree with the target search term, thus determining accurate and reliable target search results. This not only reduces the cost of information retrieval but also improves the accuracy and reliability of information retrieval.

[0075] In the above embodiments, during the retrieval process, the target search term can first be searched in the knowledge graph to obtain associated candidate triples. Then, based on the matching degree between the candidate triples and the target search term and the corresponding knowledge type, the weight of each candidate triple is determined. The target search term is then updated based on the weights, and the updated target search term is used for retrieval, thereby making the determined target search term more accurate and reliable. In actual implementation, the main entity and candidate answer entities determined by the search statement can also be searched from a preset knowledge graph database to generate a preset knowledge graph. The following section combines... Figure 2 The above process will be explained in detail.

[0076] Figure 2 This is a flowchart illustrating an information retrieval method provided in an embodiment of this disclosure, such as... Figure 2 As shown, this information retrieval method may include the following steps:

[0077] Step 201: Preprocess the obtained search statement to determine the main entity information, target entity information and core keywords corresponding to the search statement.

[0078] Among these methods, the main entity information, target entity information, and core keywords corresponding to the search statement can be determined by preprocessing the obtained search statement through SP (substance predicate) parsing, LAT (lexical answer type) identification, limitation identification, and concept identification.

[0079] It is understandable that by parsing the search query using SP (Service Point) analysis, the corresponding main entity and core keywords can be identified; by using LAT (Local Attribution-Based Adjective) identification, the target type of the search query and the target type of the main entity can be determined; by using constraint identification, the entity and core keywords can be identified; and by using concept identification, the main entity information can be determined, etc. This disclosure does not limit these aspects.

[0080] Optionally, the main entity information may include at least one of the following: the main entity identifier and the type of the main entity.

[0081] The main entity identifier uniquely represents the information, tags, etc., of the main entity. For example, the word "apple" can have multiple meanings. When referring to fruit, its corresponding identifier could be 11222; when referring to "apple" in electronic products, its corresponding main entity identifier could be 22113, and so on. This disclosure does not limit this.

[0082] In addition, the types of main entities can be divided into conceptual entities and non-conceptual entities, but this disclosure does not limit them.

[0083] It is understandable that conceptual entities can be entities with specific referents. For example, the conceptual entity "emperor" could refer to figures such as Emperor Taizong of Tang or Emperor Taizu of Song; while "Super Boy" does not have a specific referent, thus it is a non-conceptual entity. This disclosure does not impose any limitations on this.

[0084] Optionally, the target entity information may include at least one of the following: the target entity identifier and the type of the target entity.

[0085] Among them, the target entity identifier is the identifier that can uniquely represent the entity. The type of target entity can also be divided into conceptual entities and non-conceptual entities, which will not be elaborated here.

[0086] Furthermore, core keywords can be words that express actions, such as predicate verbs, or other words. For example, in the search query "Who built the Dujiangyan Irrigation System?", the core keyword is "built". This disclosure does not impose any limitations on this.

[0087] Step 202: Determine the target search terms corresponding to the search statement based on the main entity information, target entity information, and core keywords.

[0088] Optionally, if the target entity information is of a conceptual type, the target search terms can be determined based on the core keywords and the identifiers in the main entity information.

[0089] For example, if the search query is "Which emperor built the Dujiangyan Irrigation System during his reign?", where the target entity is "emperor" (a conceptual entity), the core keyword is "build", and the main entity is "Dujiangyan", then the target search terms can be determined as: Dujiangyan, build.

[0090] It should be noted that the above examples are merely illustrative and should not be construed as limiting the search statements, main entity information, target entity information, core keywords, etc., in the embodiments of this disclosure.

[0091] Optionally, if the type of the target entity information is non-conceptual, the target search terms are determined based on the identifiers in the main entity information and the target entity information.

[0092] For example, if the search query is "Wei X's year of Super Boy", and "Super Boy" is a non-conceptual entity in the target entity information, then the target search terms can be determined as: Wei X, Super Boy, based on the identifier "Wei X" in the main entity information and the identifier "Super Boy" in the target entity information.

[0093] It should be noted that the above examples are merely illustrative and should not be construed as limiting the search statements, main entity information, target entity information, core keywords, etc., in the embodiments of this disclosure.

[0094] Therefore, in this embodiment of the disclosure, target search terms can be determined in different ways according to the type in the target entity information, thereby making the determined target search terms more accurate and reliable.

[0095] Step 203: Determine the main entity and candidate answer entities corresponding to the search statement.

[0096] The candidate answer entity corresponding to the search statement can be one or more, and this disclosure does not limit this.

[0097] It is understood that candidate answer entities can be determined based on target entity information, or other methods can be used to determine candidate answer entities; this disclosure does not limit this.

[0098] For example, if the search query is "Which emperor built the Dujiangyan Irrigation System?", the main entity is "Dujiangyan", the target entity is "emperor", and the candidate answer entities can be entity information corresponding to "emperor", such as: King Zhaoxiang of Qin, Qin Shihuang, King Huiwen of Qin, etc. This publication does not limit this.

[0099] Step 204: Search the preset knowledge graph for each path from the main entity to each candidate answer entity.

[0100] It is understandable that by analyzing and processing the content in an encyclopedia knowledge base, such as through entity recognition and semantic analysis, the entities, knowledge types, and text information contained therein can be determined. Then, a pre-defined knowledge graph can be generated based on the [entity, knowledge type, text] triple. This disclosure does not limit this.

[0101] The preset knowledge graph library can store a large number of knowledge graphs, which may include various entities and relationships between entities, etc. This disclosure does not limit this.

[0102] It is understandable that the path from the main entity to the candidate answer entity searched from the preset knowledge graph can be one or multiple, and this disclosure does not limit this.

[0103] For example, in such Figure 2A In the preset knowledge graph shown, if the main entity is B and the candidate answer entity is F, then... Figure 2A The paths from the main entity "B" to the candidate answer entity "F" can be determined as: BCF, BDCF, BCEF.

[0104] It should be noted that the above-mentioned preset knowledge graph library is only partially shown and is for illustrative purposes only. It should not be taken as a limitation on the preset knowledge graph library, main entity, candidate answer entity paths, etc. in the embodiments of this disclosure.

[0105] Step 205: Generate a pre-defined knowledge graph based on each path.

[0106] Each path can contain nodes and edges, and different paths may contain common nodes. Multiple paths can be merged based on common nodes and edges to generate a predefined knowledge graph. This disclosure does not impose any limitations on this.

[0107] For example, the search query is "Which emperor built the Dujiangyan Irrigation System?", where the main entity is "Dujiangyan" and the candidate answer entity is "King Zhaoxiang of Qin". Then, a path from "Dujiangyan" to "King Zhaoxiang of Qin" can be searched in a pre-defined knowledge graph. If there are two paths, Path 1 and Path 2, between "Dujiangyan" and "King Zhaoxiang of Qin", and Path 1 contains text 1 between "Dujiangyan" and "King Zhaoxiang of Qin",... King Zhaoxiang of QinIn 256 BC, Li Bing, the governor of Shu Commandery in the Qin State, and his son, drawing on the water management experience of their predecessors, led the local people in constructing the famous Dujiangyan Irrigation System. The overall plan of Dujiangyan was to divide the Minjiang River into two streams, one of which was diverted into the Chengdu Plain. This served both to divert floods and reduce disasters, and to irrigate fields, turning a potential hazard into a benefit. In Path 2, there is text 2 between "Dujiangyan" and "Li Bing": "In 256 BC, during the Warring States period, Li Bing, the governor of Shu Commandery in the Qin State..." Li Bing The Dujiangyan Irrigation System, built by a large number of people, is located in Sichuan. Chengdu The western part of the plain, west of Dujiangyan City Min River Above, 56 kilometers from Chengdu; There is a text 3 between "Li Bing" and "King Zhaoxiang of Qin": "256 BC - 251 BC, served as the governor of Shu Commandery, and oversaw the construction of Dujiangyan, China's earliest irrigation project, near the mouth of the Minjiang River in Dujiangyan City, Sichuan Province." Since path 1 and path 2 share the common elements "Dujiangyan" and "King Zhaoxiang of Qin," it is possible to generate, based on path 1 and path 2, a... Figure 2B The preset knowledge graph shown.

[0108] It should be noted that the aforementioned preset knowledge graph library and preset knowledge graph diagram are only partially shown and are for illustrative purposes only. They should not be regarded as limitations on the preset knowledge graph library, preset knowledge graph, etc. in the embodiments of this disclosure.

[0109] Step 206: Search the preset knowledge graph based on the target search term to obtain candidate triples associated with the target search term.

[0110] Step 207: Determine the weight of each candidate triple based on the matching degree between each candidate triple and the target search term and the knowledge type corresponding to each candidate triple.

[0111] Step 208: Update the target search terms according to the weight of each candidate triplet.

[0112] Step 209: Determine the target search results based on the updated target search terms.

[0113] It should be noted that the specific content and implementation of steps 206 to 209 can be found in the descriptions of other embodiments of this disclosure, and will not be repeated here.

[0114] It is understood that the information retrieval method provided in this disclosure can be applied to any information retrieval scenario, such as dialogue question-and-answer systems, browser retrieval processes, etc., and this disclosure does not limit it.

[0115] The following is based on Figure 2C Taking this as an example, the information retrieval process provided in this disclosure will be described in detail.

[0116] For example, in a question-and-answer scenario, if the search query, or question, is "During the reign of which emperor did the Dujiangyan Irrigation System build?", the process of processing this search query to obtain the target search query, or answer, can be as follows: Figure 2C As shown.

[0117] Depend on Figure 2C It can be seen that the search query "Which emperor built Dujiangyan during his reign?" can be preprocessed first, for example, by performing "SP parsing, LAT recognition, limitation recognition, and concept recognition" to determine that the main entity is "Dujiangyan", the core verb is "build", and the target entity "emperor" is a concept entity. Then, a search can be performed in the preset knowledge graph to obtain candidate triples associated with "the construction of Dujiangyan": [256 BC; King Zhaoxiang of Qin; text 1], [256 BC; Yuzui Diversion Dam; text 1], [256 BC; Feishayan Spillway; text 1], [256 BC; Li Bing; text 2], [256 BC; Chengdu; text 2], [256 BC; Li Bing; text 2].

[0118] Among them, text 1 is " King Zhaoxiang of Qin In 256 BC, Li Bing, the governor of Shu Commandery in the Qin State, and his son, drawing on the water management experience of their predecessors, led the local people in constructing the famous Dujiangyan Irrigation System. The overall plan of Dujiangyan was to divide the Minjiang River into two streams, one of which was diverted into the Chengdu Plain. This served both to divert floods and reduce disasters, and to irrigate farmland, turning a potential hazard into a beneficial one. The main project included... Fish Mouth Dividing Embankment , Feisha Weir Spillway "And the water inlet of the bottle mouth." Text 2 is "In 256 BC, the governor of Shu County of the Qin State during the Warring States period." Li Bing The Dujiangyan Irrigation System, built by a large number of people, is located in Sichuan. Chengdu The western part of the plain, west of Dujiangyan City Minjiang River It is located 56 kilometers from Chengdu.

[0119] Next, the conceptual entity "Emperor" can be validated for conceptual knowledge, and the validation result is "Emperor - King Zhaoxiang of Qin". Then, the time knowledge can be validated, and the validation result is "256 BC - Reign of King Zhaoxiang of Qin". After that, inference relevance calculation can be performed, such as "Li Bing -> King Zhaoxiang of Qin", "Li Bing -> Qin Shi Huang", "...", and finally the target search result "King Zhaoxiang of Qin" can be determined.

[0120] It should be noted that the above examples are merely illustrative and should not be construed as limiting the information retrieval methods and processes in the embodiments of this disclosure.

[0121] In this embodiment, the acquired search statement can be preprocessed to determine the main entity information, target entity information, and core keywords corresponding to the search statement. Then, based on the main entity information, target entity information, and core keywords, the target search term corresponding to the search statement can be determined. Next, the main entity and candidate answer entities corresponding to the search statement can be determined first. Each path from the main entity to each candidate answer entity can be searched from a preset knowledge graph library. Then, a preset knowledge graph can be generated based on each path. The preset knowledge graph can be searched based on the target search term to obtain candidate triples associated with the target search term. Then, based on the matching degree between each candidate triple and the target search term and the knowledge type corresponding to each candidate triple, the weight of each candidate triple can be determined. Then, based on the weight of each candidate triple, the target search term can be updated. Finally, based on the updated target search term, the target search result can be determined. Therefore, by using the main entity information, target entity information and core keywords corresponding to the search statement, a relatively accurate target search term can be determined. Then, by combining the generated pre-set knowledge graph, related candidate triples can be obtained. The target search term is then updated based on the candidate triples, which makes the determined target search results more accurate and reliable, further improving the accuracy and reliability of information retrieval.

[0122] The above embodiments can determine relatively accurate target search terms by using the main entity information, target entity information, and core keywords corresponding to the search statement. Then, by combining this with a pre-generated knowledge graph, associated candidate triples can be obtained. The target search terms are updated based on these candidate triples, thereby making the determined target search results more accurate and reliable. In one possible implementation, the weight of any candidate triple can be determined based on the initial weight and association weight of any candidate triple in the pre-generated knowledge graph. The following will combine... Figure 3 The above process will be explained.

[0123] Figure 3 This is a flowchart illustrating an information retrieval method provided in an embodiment of this disclosure, such as... Figure 3 As shown, this information retrieval method may include the following steps:

[0124] Step 301: Determine the target search terms corresponding to the search statement.

[0125] Step 302: Search the preset knowledge graph based on the target search term to obtain candidate triples associated with the target search term.

[0126] It should be noted that the specific content and implementation of steps 301 and 302 can be found in the descriptions of other embodiments of this disclosure, and will not be repeated here.

[0127] Step 303: Determine the number of paths in any candidate triple contained in the preset knowledge graph, and the text source in any candidate triple.

[0128] For example, in such Figure 3A In the preset knowledge graph shown, if any candidate triple is "CD" in the graph, then by Figure 3A It can be seen that in the preset knowledge graph, the number of paths containing any candidate triple "CD" is 4.

[0129] It should be noted that the above examples are merely illustrative and should not be construed as limiting the pre-defined knowledge graph, any candidate triple, etc., in the embodiments of this disclosure.

[0130] In addition, since any candidate triple contains corresponding text, the source of the text in any candidate triple can be determined based on that text.

[0131] The source of the text can be various, such as an article, a book, an encyclopedia, etc. This publication does not limit this.

[0132] For example, the text in any candidate triple is "256 BC, the governor of Shu Commandery of the Qin State during the Warring States period". Li Bing The Dujiangyan Irrigation System, built by a large number of people, is located in Sichuan. Chengdu The western part of the plain, west of Dujiangyan City Minjiang River The phrase "56 kilometers from Chengdu" can be searched to determine the source of the text, for example, it can be determined that the source is an encyclopedia knowledge base.

[0133] It should be noted that the above examples are merely illustrative and should not be construed as limiting the text, text source, etc., in any candidate triplet of the embodiments of this disclosure.

[0134] Furthermore, the text source in any candidate triple can be one or more, and this disclosure does not limit this.

[0135] Step 304: Determine the initial weight of any candidate triple based on the number of paths containing any candidate triple, the text source in any candidate triple, and the matching degree between any candidate triple and the target search term.

[0136] It is understandable that the more paths containing any candidate triple, the greater the initial weight of that candidate triple; or, the more text sources in any candidate triple, the greater the initial weight of that candidate triple; or, the higher the match between any candidate triple and the target search term, the greater the initial weight of that candidate triple. This disclosure does not limit this.

[0137] For example, if the number of paths containing any candidate triplet 1 is 3, and the text source in any candidate triplet 1 is text 1 and text 2, then the match degree between any candidate triplet 1 and the target search term is 0.8. For any candidate triplet 2, if the number of paths containing any candidate triplet 2 is 1, and the text source in any candidate triplet 2 is text 3, then the match degree between any candidate triplet 2 and the target search term is 0.5. Therefore, it can be determined that the initial weight of any candidate triplet 1 is slightly larger, such as 0.5, 0.6, etc.; and the initial weight of any candidate triplet 2 is determined to be slightly smaller, such as 0.3, 0.2, etc.

[0138] It should be noted that the above examples are merely illustrative and should not be taken as limitations on the method of determining the initial weight of any candidate triple in the embodiments of this disclosure.

[0139] Step 305: Determine the association weight corresponding to any candidate triple based on the association relationships between the remaining candidate triples with the same knowledge type as any candidate triple and any candidate triple.

[0140] For example, if the knowledge type of any candidate triple 1 is "2007", the knowledge type of candidate triple 2 among the other candidate triples is "2007", and the knowledge type of candidate triple 3 is "February 2007", and "February 2007" can be used to deduce the year "2007", then the association weight corresponding to any candidate triple 1 can be determined to be 0.3, or 0.4, etc.

[0141] Alternatively, if the knowledge type of any candidate triple 1 is "2007", the knowledge type of candidate triple 2 among the other candidate triples is "2007", the knowledge type of candidate triple 3 is "February 2007", and the knowledge type of candidate triple 4 is "June 21, 2007", where "February 2007" and "June 21, 2007" can both be deduced to belong to the year "2007", then it can be determined that the association weight corresponding to any candidate triple 1 is slightly larger, such as 0.7, 0.65, etc.

[0142] Alternatively, if the knowledge type of any candidate triple 1 is "2007", the knowledge type of candidate triple 2 among the other candidate triples is "2006", and the knowledge type of candidate triple 3 is "February 2008", then the association weight corresponding to any candidate triple 1 can be determined to be 0, or it can be a small value, such as 0.01, 0.005, etc.

[0143] It should be noted that the above examples are merely illustrative and should not be construed as limiting the methods for determining the association weights corresponding to any candidate triples in the embodiments of this disclosure.

[0144] Optionally, if the value of the knowledge type in the first candidate triplet is a subset of the value of the knowledge type in any candidate triplet, then the association weight corresponding to any candidate triplet is determined to include the initial weight corresponding to the first candidate triplet.

[0145] For example, if the knowledge type value in any candidate triple is "2007" and the knowledge type value in the first candidate triple is "June 2007", since "2007" can include "June 2007", meaning "June 2007" in the first candidate triple is a subset of "2007" in any candidate triple, then the association weight corresponding to any candidate triple can be determined to include the initial weight corresponding to the first candidate triple. For example, if the initial weight of the first candidate triple is 0.2 and the initial weight of any candidate triple is 0.3, then the association weight of that candidate triple can be (0.2 + 0.3), which is 0.5.

[0146] It should be noted that the above examples are merely illustrative and should not be construed as limiting the methods for determining the association weights corresponding to any candidate triples in the embodiments of this disclosure.

[0147] Step 306: Determine the weight of any candidate triple based on the initial weight and associated weight.

[0148] It is understandable that the initial weight and associated weight of any candidate triple can be added together to obtain the weight of that candidate triple. Alternatively, the initial weight and associated weight can be proportionally combined to obtain the weight of that candidate triple, and so on. This disclosure does not limit this.

[0149] Optionally, the weight of any candidate triple can be expressed by the following formula (1):

[0150]

[0151] in, This represents the association weight of x. Let x be the initial weight.

[0152] For example, the knowledge type of candidate triple 1 is 2007, the knowledge type of candidate triple 2 is June 2007, and the knowledge type of candidate triple 3 is 2006, where "June 2007" is a subset of "2007". Therefore, according to formula (1), the association weight of candidate triple 1 is... That is, the initial weights corresponding to candidate triple 2 and candidate triple 1 are... The sum of the two is the weight of candidate triple 1. This disclosure does not limit this.

[0153] Therefore, in this embodiment of the present disclosure, when determining the weight of any candidate triple, the initial weight of any candidate triple can be determined first based on the number of paths containing any candidate triple in the preset knowledge graph, the text source of any candidate triple, and the matching degree between any candidate triple and the target search term. Then, the association weight of any candidate triple can be determined based on the knowledge type of any candidate triple. This makes the determined weight of any candidate triple more accurate and reliable, providing conditions for improving the reliability and accuracy of information retrieval.

[0154] Step 307: Combine the entities in any candidate triple with a weight greater than the threshold with the target search term to generate a new target search term.

[0155] The threshold can be a pre-set value, such as 0.6, 0.75, etc., and this disclosure does not limit it.

[0156] For example, if the threshold is set to 0.8 and the weight of any candidate triple is 0.85, which is greater than the threshold, then if the entity in any candidate triple is "Minjiang River" and the target search term is "Dujiangyan, construction", then the generated new target search term can be "Minjiang River, Dujiangyan, construction".

[0157] It should be noted that the above examples are merely illustrative and should not be construed as limiting the threshold, entities in any candidate triplet, target search terms, etc., in the embodiments of this disclosure.

[0158] Step 308: Determine the target search results based on the updated target search terms.

[0159] It should be noted that the specific content and implementation method of step 308 can be referred to the descriptions of other embodiments of this disclosure, and will not be repeated here.

[0160] In this embodiment, the target search term corresponding to the search statement can be determined first. Then, a preset knowledge graph can be searched based on the target search term to obtain candidate triples associated with the target search term. Next, the number of paths containing any candidate triple in the preset knowledge graph and the text source in any candidate triple can be determined. Then, the initial weight of any candidate triple can be determined based on the number of paths containing any candidate triple, the text source in any candidate triple, and the matching degree between any candidate triple and the target search term. Then, the association weight corresponding to any candidate triple can be determined based on the association relationship between other candidate triples with the same knowledge type as any candidate triple and any candidate triple. Then, the weight of any candidate triple can be determined based on the initial weight and the association weight. Then, the entities in any candidate triple with a weight greater than a threshold can be combined with the target search term to generate a new target search term. Finally, the target search result can be determined based on the updated target search term. Therefore, when determining the weight of any candidate triple, the preset knowledge graph and the knowledge type of any candidate triple are fully considered, thereby improving the accuracy and reliability of weight determination. Consequently, when updating the target search terms based on the weight of the candidate triple, the updated target search terms can be more accurate and reliable, further improving the accuracy and reliability of information retrieval.

[0161] Understandably, after identifying the target search term, the confidence level of the target search result can be determined based on the historical data corresponding to that term. Only if the confidence level meets a threshold condition will the target search result be displayed. The following section combines... Figure 4 The above process will be explained in detail.

[0162] Figure 4 This is a flowchart illustrating an information retrieval method provided in an embodiment of this disclosure, such as... Figure 4 As shown, this information retrieval method may include the following steps:

[0163] Step 401: Determine the target search term corresponding to the search statement.

[0164] Step 402: Search the preset knowledge graph based on the target search term to obtain candidate triples associated with the target search term.

[0165] Step 403: Determine the weight of each candidate triple based on the matching degree between each candidate triple and the target search term and the knowledge type corresponding to each candidate triple.

[0166] Step 404: Update the target search terms according to the weight of each candidate triplet.

[0167] It should be noted that the specific content and implementation of steps 401 to 404 can be found in the descriptions of other embodiments of this disclosure, and will not be repeated here.

[0168] Step 405: Determine the target search results based on the updated target search terms.

[0169] Optionally, the entity key-value pair library can be searched based on the updated target search terms. If the entity key-value pair library contains a target entity corresponding to the updated target search terms, the target entity can be identified as the target search result.

[0170] The entity key-value pair library can be pre-defined and may contain numerous entity key-value pairs. For example, it could be "the first emperor of the Qin Dynasty - Ying Zheng" or "the first AI program to defeat the world Go champion - AlphaGo," etc. This disclosure does not impose any limitations on this.

[0171] For example, if the search query is "Wei X's year Super Boy champion", and the target search term is "Wei X Super Boy champion", the updated target search term determined through the above processing is "2007 Super Boy champion". Then, you can search for the entity corresponding to "2007 Super Boy champion" in the entity key-value pair database. If the retrieved entity key-value pair is "2007 Super Boy champion - Chen XX", then you can determine that "Chen XX" is the target search result.

[0172] It should be noted that the above examples are merely illustrative and should not be construed as limiting the entity key-value pairs, target search terms, etc., in the embodiments of this disclosure.

[0173] Alternatively, if the entity key-value pair library does not contain the target entity corresponding to the updated target search term, the preset knowledge graph is searched based on the updated target search term to determine the target search result corresponding to the updated target search term.

[0174] For example, if the search query is "During which emperor's reign was the Dujiangyan Irrigation System built?", and the updated target query is "Dujiangyan Irrigation System was built in 256 BC", but the entity key-value pair database does not contain a target entity corresponding to the updated target query, then a search can be performed on the preset knowledge graph based on the updated target query "Dujiangyan Irrigation System was built in 256 BC". If the preset knowledge graph is like... Figure 2A As shown, by searching this knowledge graph, if the search result is "the 51st year of King Zhaoxiang of Qin (256 BC)," the corresponding target search result can be determined to be "King Zhaoxiang of Qin" by verifying the "time knowledge," that is, the reign time of King Zhaoxiang of Qin.

[0175] It should be noted that the above examples are merely illustrative and should not be construed as limiting the updated target search terms, target search results, etc., in the embodiments of this disclosure.

[0176] Therefore, in this embodiment of the disclosure, after determining the updated target search terms, the target search results can be obtained by searching the entity key-value pair library or the preset knowledge graph, thereby making the determined target search results more accurate and reliable.

[0177] Step 406: Obtain the historical search results display page data corresponding to the updated target search term.

[0178] The historical search results display page data can be data on historical search results corresponding to the target search term. For example, it can be the historical search count, historical search time, etc., and this disclosure does not limit it.

[0179] In addition, the data displayed on the historical search results page can be in table format, text format, chart format, etc., and this disclosure does not limit it in this way.

[0180] Step 407: Determine the frequency of the target search result appearing on the historical search results display page.

[0181] The frequency of a target search result appearing on a historical search results page can be determined by searching that page. This disclosure does not limit this aspect.

[0182] Step 408: Determine the confidence level of the target search result based on the frequency of occurrence and the search result operation data in the historical search result display page data.

[0183] Among them, the search result operation data can be data that performs operations on the search results, such as click operations, browsing operations, etc., and this disclosure does not limit it.

[0184] For example, if the search query is "Which emperor built the Dujiangyan Irrigation System?", and the target search term is determined to be "King Zhaoxiang of Qin", and if the frequency of "King Zhaoxiang of Qin" in the historical search results page is 10,000, with 9,900 corresponding "clicks" and "views", while "Qin Shi Huang" appears 111 times, with 2 "clicks" and 1 "view", then the confidence level of the target search result is relatively high, such as 0.9, 0.88, etc., and this disclosure does not limit this.

[0185] It should be noted that the above examples are merely illustrative and should not be construed as limiting the frequency of occurrence of target search results, search result operation data, and confidence level in the embodiments of this disclosure.

[0186] Step 409: If the confidence level is greater than the threshold, display the target search result on the search results display page.

[0187] The threshold can be a pre-set value, such as 0.9, 0.88, etc., and this disclosure does not limit it.

[0188] Understandably, when the confidence level is greater than the threshold, the target search result can be considered to have high reliability, and it can be displayed on the search results page. When the confidence level is less than or equal to the threshold, the target search result can be considered to have low reliability. This disclosure does not impose any limitations in this regard.

[0189] The following is combined Figure 4A This document explains the information retrieval process provided in this disclosure.

[0190] Depend on Figure 4A It can be seen that the search query "Which emperor built Dujiangyan during his reign?" can be preprocessed first, for example, by performing "SP parsing, LAT recognition, limitation recognition, and concept recognition". This can determine that the main entity is "Dujiangyan", the core verb is "build", and the target entity "emperor" is a conceptual entity. Then, a search can be performed in a pre-defined knowledge graph to obtain candidate triples associated with "Dujiangyan construction". Next, the conceptual entity "emperor" can be validated for conceptual knowledge, resulting in "emperor - King Zhaoxiang of Qin". Then, a time-related knowledge validation can be performed, resulting in "256 BC - the reign of King Zhaoxiang of Qin". Finally, inference relevance calculations can be performed, such as "Li Bing -> King Zhaoxiang of Qin, Li Bing -> Qin Shi Huang, ...". Then, historical search results display page data can be obtained. Based on the frequency of the target search result appearing on the historical search results display page and the search result operation data, the confidence level of "King Zhaoxiang of Qin" can be determined. If the confidence level of the target search result is greater than the threshold, the target search result can be displayed on the search results display page. For example, if the confidence level is greater than the threshold, the target search result is determined to be "King Zhaoxiang of Qin".

[0191] It should be noted that the above examples are merely illustrative and should not be construed as limiting the information retrieval process or other aspects of the embodiments of this disclosure.

[0192] It is understood that the information retrieval method provided in this disclosure can be applied to any information retrieval framework. The following example illustrates this. Figure 4B The information retrieval framework shown is used as an example for illustration.

[0193] Depend on Figure 4BAs can be seen, the search query is first preprocessed using SP parsing, LAT recognition, constraint recognition, and concept recognition to determine the main entity information, target entity information, core keywords, and target search terms. Then, the entity retrieval module searches the entity database for the input main entity information to output the main entity identifier. The text extraction module uses a pre-trained model to process the search query to determine the corresponding main entity and candidate answer entities. The text retrieval module can perform a search based on the entity information. The knowledge reasoning module may include: a graph construction unit, a knowledge tuple extraction unit, a tuple aggregation and scoring unit, and a multi-hop / implicit knowledge reasoning unit. The graph construction unit can generate a preset knowledge graph based on the main entity, candidate answer entities, and a preset knowledge graph database. The knowledge tuple extraction unit can search the preset knowledge graph based on the target search terms to obtain associated candidate triples. The tuple aggregation and scoring unit can determine the weight of each candidate triple based on its initial weight and association weight. The multi-hop / implicit knowledge reasoning unit can update the target search terms based on the weight of each candidate triple to determine the target search result, i.e., the answer.

[0194] It should be noted that the above examples are merely illustrative and should not be construed as limiting the information retrieval framework or other aspects of the embodiments of this disclosure.

[0195] In this embodiment, the target search term corresponding to the search query is first determined. Then, a preset knowledge graph is searched based on the target search term to obtain candidate triples associated with the target search term. The weight of each candidate triple is determined according to its matching degree with the target search term and the knowledge type it corresponds to. The target search term is then updated based on its weight. Historical search result display page data corresponding to the updated target search term is then obtained, and the frequency of the target search result appearing on the historical search result display page is determined. The confidence level of the target search result is then determined based on the frequency of appearance and the search result operation data in the historical search result display page data. If the confidence level is greater than a threshold, the target search result is displayed on the search result display page. Therefore, after determining the target search result, the confidence level of the target search result can be determined based on historical search data. If the confidence level is greater than a threshold, the target search result is then displayed, thereby improving the reliability and accuracy of the target search result.

[0196] To implement the above embodiments, this disclosure also proposes an information retrieval device.

[0197] Figure 5 This is a schematic diagram of the structure of an information retrieval device provided in an embodiment of the present disclosure.

[0198] like Figure 5 As shown, the information retrieval device 500 includes: a first determining module 510, an acquiring module 520, a second determining module 530, an updating module 540, and a third determining module 550.

[0199] The first determining module 510 is used to determine the target search term corresponding to the search statement.

[0200] The acquisition module 520 is used to search a preset knowledge graph based on the target search term in order to obtain candidate triples associated with the target search term.

[0201] The second determining module 530 is used to determine the weight of each candidate triple based on the matching degree between each candidate triple and the target search term and the knowledge type corresponding to each candidate triple.

[0202] The update module 540 is used to update the target search term according to the weight of each candidate triplet.

[0203] The third determination module 550 is used to determine the target search results based on the updated target search terms.

[0204] Optionally, the first determining module includes:

[0205] The first determining unit is used to preprocess the acquired search statement to determine the main entity information, target entity information and core keywords corresponding to the search statement;

[0206] The second determining unit is used to determine the target search term corresponding to the search statement based on the main entity information, target entity information and core keywords.

[0207] Optionally, the main entity information includes at least one of the following: main entity identifier and main entity type;

[0208] The target entity information includes at least one of the following: target entity identifier and target entity type.

[0209] Optionally, the second determining unit is specifically used for:

[0210] If the type of the target entity information is a concept, the target search term is determined based on the core keywords and the identifier in the main entity information;

[0211] Alternatively, if the type of the target entity information is non-conceptual, the target search term is determined based on the identifier in the main entity information and the identifier in the target entity information.

[0212] Optionally, the first determining module is further configured to:

[0213] Determine the main entity and candidate answer entities corresponding to the search query;

[0214] Search for each path from the main entity to each of the candidate answer entities in the preset knowledge graph database;

[0215] Based on each path, the preset knowledge graph is generated.

[0216] Optionally, the second determining module includes:

[0217] The third determining unit is used to determine the number of paths in any candidate triple contained in the preset knowledge graph, and the text source in any candidate triple;

[0218] The fourth determining unit is used to determine the initial weight of any candidate triple based on the number of paths containing any candidate triple, the text source in any candidate triple, and the matching degree between any candidate triple and the target search term.

[0219] The fifth determining unit is used to determine the association weight corresponding to any candidate triple based on the association relationship between the remaining candidate triples with the same knowledge type as the candidate triple and the candidate triple.

[0220] The sixth determining unit is used to determine the weight of any candidate triple based on the initial weight and the associated weight.

[0221] Optionally, the fifth determining unit is specifically used for:

[0222] If the value of the knowledge type in the first candidate triplet is a subset of the value of the knowledge type in any candidate triplet, then the association weight corresponding to any candidate triplet is determined to include the initial weight corresponding to the first candidate triplet.

[0223] Optionally, the update module is specifically used for:

[0224] The entities in any candidate triple with a weight greater than the threshold are combined with the target search term to generate a new target search term.

[0225] Optionally, the third determining module is specifically used for:

[0226] Based on the updated target search terms, the entity key-value pair library is searched. If the entity key-value pair library contains a target entity corresponding to the updated target search terms, the target entity is confirmed as the target search result.

[0227] If the entity key-value pair library does not contain a target entity corresponding to the updated target search term, the preset knowledge graph is searched based on the updated target search term to determine the target search result corresponding to the updated target search term.

[0228] Optionally, the first determining module is further configured to:

[0229] Obtain the historical search results display page data corresponding to the updated target search term;

[0230] Determine the frequency of occurrence of the target search result in the historical search results display page;

[0231] The confidence level of the target search result is determined based on the frequency of occurrence and the search result operation data in the historical search result display page data.

[0232] If the confidence level is greater than the threshold, the target search result is displayed on the search results display page.

[0233] The functions and specific implementation principles of the modules described in this embodiment can be found in the above method embodiments, and will not be repeated here.

[0234] The information retrieval device of this disclosure first determines the target search term corresponding to the search statement, then searches a preset knowledge graph based on the target search term to obtain candidate triples associated with the target search term, and then determines the weight of each candidate triple based on the matching degree between each candidate triple and the target search term and the knowledge type corresponding to each candidate triple, and then updates the target search term based on the weight of each candidate triple, and determines the target retrieval result based on the updated target search term. Therefore, in the retrieval process, the target search term can be searched in the knowledge graph first to obtain associated candidate triples, and then the target search term can be updated based on the knowledge type corresponding to the candidate triple and the matching degree with the target search term, thus determining accurate and reliable target retrieval results, thereby reducing the cost of information retrieval and improving the accuracy and reliability of information retrieval.

[0235] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0236] Figure 6A schematic block diagram of an example electronic device 600 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0237] like Figure 6 As shown, device 600 includes a computing unit 601, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 602 or a computer program loaded from storage unit 608 into random access memory (RAM) 603. RAM 603 may also store various programs and data required for the operation of device 600. The computing unit 601, ROM 602, and RAM 603 are interconnected via bus 604. Input / output (I / O) interface 605 is also connected to bus 604.

[0238] Multiple components in device 600 are connected to I / O interface 605, including: input unit 606, such as keyboard, mouse, etc.; output unit 607, such as various types of monitors, speakers, etc.; storage unit 608, such as disk, optical disk, etc.; and communication unit 609, such as network card, modem, wireless transceiver, etc. Communication unit 609 allows device 600 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0239] The computing unit 601 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 601 performs the various methods and processes described above, such as information retrieval methods. For example, in some embodiments, the information retrieval method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 608. In some embodiments, part or all of the computer program may be loaded and / or installed on device 600 via ROM 602 and / or communication unit 609. When the computer program is loaded into RAM 603 and executed by the computing unit 601, one or more steps of the information retrieval method described above may be performed. Alternatively, in other embodiments, the computing unit 601 may be configured to perform the information retrieval method by any other suitable means (e.g., by means of firmware).

[0240] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0241] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0242] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0243] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0244] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), the Internet, and blockchain networks.

[0245] Computer systems can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. A server can be a cloud server, also known as a cloud computing server or cloud host, a hosting product within the cloud computing service ecosystem, addressing the shortcomings of traditional physical hosts and VPS (Virtual Private Server, or simply "VPS") services, such as high management difficulty and weak business scalability. Servers can also be servers for distributed systems or servers incorporating blockchain technology.

[0246] The technical solution disclosed herein can first determine the target search terms corresponding to the search query, then search a preset knowledge graph based on the target search terms to obtain candidate triples associated with the target search terms, and then determine the weight of each candidate triple based on the matching degree between each candidate triple and the target search terms and the knowledge type corresponding to each candidate triple. Finally, the target search terms are updated based on the weights of each candidate triple, and the target search results are determined based on the updated target search terms. Therefore, in the retrieval process, the target search terms can be searched in the knowledge graph first to obtain associated candidate triples, and then the target search terms can be updated based on the knowledge type corresponding to the candidate triples and the matching degree with the target search terms, thus determining accurate and reliable target search results. This reduces the cost of information retrieval and improves the accuracy and reliability of information retrieval.

[0247] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0248] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. An information retrieval method, comprising: Determine the target search terms corresponding to the search query; The preset knowledge graph is searched based on the target search term to obtain candidate triples associated with the target search term; The weight of each candidate triple is determined based on the matching degree between each candidate triple and the target search term and the knowledge type corresponding to each candidate triple; The target search terms are updated according to the weight of each candidate triple; Based on the updated target search terms, determine the target search results; The step of determining the weight of each candidate triple based on the matching degree between each candidate triple and the target search term and the knowledge type corresponding to each candidate triple includes: Determine the number of paths in any candidate triple contained in the preset knowledge graph, and the text source in any candidate triple; The initial weight of any candidate triple is determined based on the number of paths in any candidate triple contained in the preset knowledge graph, the text source in any candidate triple, and the matching degree between any candidate triple and the target search term. Based on the association relationships between the remaining candidate triples with the same knowledge type as the candidate triple and the candidate triple, the association weight corresponding to the candidate triple is determined; The weight of any candidate triple is determined based on the initial weight and the associated weight.

2. The method as described in claim 1, wherein, The determination of the target search term corresponding to the search statement includes: The obtained search statements are preprocessed to determine the main entity information, target entity information and core keywords corresponding to the search statements; Based on the main entity information, target entity information, and core keywords, the target search terms corresponding to the search statement are determined.

3. The method as described in claim 2, wherein, The main entity information includes at least one of the following: main entity identifier and main entity type; The target entity information includes at least one of the following: target entity identifier and target entity type.

4. The method of claim 3, wherein, The step of determining the target search term corresponding to the search statement based on the main entity information, target entity information, and core keywords includes: If the type of the target entity information is a concept, the target search term is determined based on the core keywords and the identifier in the main entity information; Alternatively, if the type of the target entity information is non-conceptual, the target search term is determined based on the identifier in the main entity information and the identifier in the target entity information.

5. The method of claim 1, wherein, Before performing the search on the preset knowledge graph based on the target search term, the method further includes: Determine the main entity and candidate answer entities corresponding to the search query; Search for each path from the main entity to each of the candidate answer entities in the preset knowledge graph database; Based on each path, the preset knowledge graph is generated.

6. The method of claim 1, wherein, The step of determining the association weight corresponding to any candidate triplet based on the association relationships between other candidate triplets of the same knowledge type as the candidate triplet and the candidate triplet includes: If the value of the knowledge type in the first candidate triplet is a subset of the value of the knowledge type in any candidate triplet, then the association weight corresponding to any candidate triplet is determined to include the initial weight corresponding to the first candidate triplet.

7. The method of claim 1, wherein, The step of updating the target search term according to the weight of each candidate triplet includes: The entities in any candidate triple with a weight greater than the threshold are combined with the target search term to generate a new target search term.

8. The method as described in any one of claims 1-7, wherein, The process of determining the target search results based on the updated target search terms includes: Based on the updated target search terms, the entity key-value pair library is searched. If the entity key-value pair library contains a target entity corresponding to the updated target search terms, the target entity is confirmed as the target search result. If the entity key-value pair library does not contain a target entity corresponding to the updated target search term, the preset knowledge graph is searched based on the updated target search term to determine the target search result corresponding to the updated target search term.

9. The method as described in any one of claims 1-7, wherein, After determining the target search results based on the updated target search terms, the process further includes: Obtain the historical search results display page data corresponding to the updated target search term; Determine the frequency of occurrence of the target search result in the historical search results display page; The confidence level of the target search result is determined based on the frequency of occurrence and the search result operation data in the historical search result display page data; If the confidence level is greater than the threshold, the target search result will be displayed on the search results page.

10. An information retrieval device, comprising: The first determining module is used to determine the target search terms corresponding to the search statement; The acquisition module is used to search a preset knowledge graph based on the target search term in order to obtain candidate triples associated with the target search term; The second determining module is used to determine the weight of each candidate triple based on the matching degree between each candidate triple and the target search term and the knowledge type corresponding to each candidate triple; An update module is used to update the target search term according to the weight of each candidate triple; The third determination module is used to determine the target search results based on the updated target search terms; The second determining module includes: The third determining unit is used to determine the number of paths in any candidate triple contained in the preset knowledge graph, and the text source in any candidate triple; The fourth determining unit is used to determine the initial weight of any candidate triple based on the number of paths in any candidate triple contained in the preset knowledge graph, the text source in any candidate triple, and the matching degree between any candidate triple and the target search term. The fifth determining unit is used to determine the association weight corresponding to any candidate triple based on the association relationship between the remaining candidate triples with the same knowledge type as the candidate triple and the candidate triple. The sixth determining unit is used to determine the weight of any candidate triple based on the initial weight and the associated weight.

11. The apparatus of claim 10, wherein, The first determining module includes: The first determining unit is used to preprocess the acquired search statement to determine the main entity information, target entity information and core keywords corresponding to the search statement; The second determining unit is used to determine the target search term corresponding to the search statement based on the main entity information, target entity information, and core keywords.

12. The apparatus of claim 11, wherein, The main entity information includes at least one of the following: main entity identifier and main entity type; The target entity information includes at least one of the following: target entity identifier and target entity type.

13. The apparatus of claim 12, wherein, The second determining unit is specifically used for: If the type of the target entity information is a concept, the target search term is determined based on the core keywords and the identifier in the main entity information; Alternatively, if the type of the target entity information is non-conceptual, the target search term is determined based on the identifier in the main entity information and the identifier in the target entity information.

14. The apparatus of claim 10, wherein, The first determining module is further configured to: Determine the main entity and candidate answer entities corresponding to the search query; Search for each path from the main entity to each of the candidate answer entities in the preset knowledge graph database; Based on each path, the preset knowledge graph is generated.

15. The apparatus of claim 10, wherein, The fifth determining unit is specifically used for: If the value of the knowledge type in the first candidate triplet is a subset of the value of the knowledge type in any candidate triplet, then the association weight corresponding to any candidate triplet is determined to include the initial weight corresponding to the first candidate triplet.

16. The apparatus of claim 10, wherein, The update module is specifically used for: The entities in any candidate triple with a weight greater than the threshold are combined with the target search term to generate a new target search term.

17. The apparatus according to any one of claims 10-16, wherein, The third determining module is specifically used for: Based on the updated target search terms, the entity key-value pair library is searched. If the entity key-value pair library contains a target entity corresponding to the updated target search terms, the target entity is confirmed as the target search result. If the entity key-value pair library does not contain a target entity corresponding to the updated target search term, the preset knowledge graph is searched based on the updated target search term to determine the target search result corresponding to the updated target search term.

18. The apparatus according to any one of claims 10-16, wherein, The first determining module is further configured to: Obtain the historical search results display page data corresponding to the updated target search term; Determine the frequency of occurrence of the target search result in the historical search results display page; The confidence level of the target search result is determined based on the frequency of occurrence and the search result operation data in the historical search result display page data; If the confidence level is greater than the threshold, the target search result will be displayed on the search results page.

19. An electronic device comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-9.

20. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-9.

21. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-9.

Images