WO2020143319A1 - 知识图谱补全方法、装置、计算机设备及存储介质 - Google Patents
知识图谱补全方法、装置、计算机设备及存储介质 Download PDFInfo
- Publication number
- WO2020143319A1 WO2020143319A1 PCT/CN2019/118160 CN2019118160W WO2020143319A1 WO 2020143319 A1 WO2020143319 A1 WO 2020143319A1 CN 2019118160 W CN2019118160 W CN 2019118160W WO 2020143319 A1 WO2020143319 A1 WO 2020143319A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- entity
- knowledge graph
- named
- tuple
- knowledge
- Prior art date
- Legal status (The legal status 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 status listed.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/26—Visual data mining; Browsing structured data
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
Definitions
- the present application relates to the field of information processing technology, and in particular, to a knowledge graph completion method, device, computer equipment, and storage medium.
- the knowledge graph is a knowledge cluster organized in the form of a graph in knowledge engineering. It consists of different types of entities as nodes and relationships as edges connecting nodes.
- an entity refers to an objective object in the real world, or an abstract concept in human thought, and a relationship describes the actual relationship between two entities.
- entity types include people, events, organizations, places, etc., and the types of relationships between them are also very diverse. Different entity types have different relationships. For example, for the relationship between people and entities, the common relationship is the relationship between relatives and friends; for the relationship between people and organizations, the common relationship is the work unit, graduated schools, etc. The relationship between these known entities is relatively sparse in the original knowledge graph, but in fact there are a large number of implicit relationships between entities.
- the inventor realizes that in the existing mechanism, when constructing and completing the knowledge graph, all the acquired entities in the data source will be analyzed, and then the relationship between all entities and entity attributes will be established. In this way, although the constructed knowledge graph can cover a wide range, for a batch of samples, not only need to know the data of each sample, but also need to know the corresponding category label, which is very dependent on the data source containing the label. higher cost.
- Embodiments of the present application can provide a knowledge graph completion method, device, computer equipment, and storage medium that reduce dependence on a data source containing an annotation.
- an embodiment of the present application provides a knowledge graph completion method, including the following steps: acquiring knowledge information to be processed; and extracting named entities from the knowledge information to be processed according to a preset first extraction rule And association relationship; generating entity tuples according to the named entities and association relationships; supplementing the entity tuples to the knowledge graph.
- an embodiment of the present application further provides a knowledge graph completion device, including: an acquisition module for acquiring knowledge information to be processed; and an extraction module for extracting from the Named entities and association relationships are extracted from the knowledge information to be processed; a processing module is used to generate entity tuples based on the named entities and association relationships; an execution module is used to supplement the entity tuples into the knowledge graph.
- embodiments of the present application further provide a computer device, including a memory and a processor, and the memory stores computer-readable instructions.
- the processor executes a knowledge graph completion method.
- the knowledge graph completion method includes the following steps: acquiring knowledge information to be processed; and extracting named entities from the knowledge information to be processed according to a preset first extraction rule And association relationship; generating entity tuples according to the named entities and association relationships; supplementing the entity tuples to the knowledge graph.
- embodiments of the present application also provide a computer storage medium on which a computer program is stored.
- the processor is caused to perform a knowledge graph completion method.
- the knowledge graph completion method includes the following steps: acquiring knowledge information to be processed; extracting named entities and association relationships from the knowledge information to be processed according to a preset first extraction rule; generating based on the named entities and association relationships Entity tuple; supplement the entity tuple to the knowledge graph.
- FIG. 1 is a schematic diagram of a basic process of a knowledge graph completion method according to an embodiment of this application;
- FIG. 2 is a schematic flowchart of extracting named entities and associated relationships according to an embodiment of the present application
- FIG. 3 is a schematic flowchart of linking entity tuples according to an embodiment of the present application.
- FIG. 4 is a schematic diagram of a process of classifying named entities and supplementing relationships according to an embodiment of the present application
- FIG. 5 is a schematic diagram of a process of supplementing attribute tuples according to an embodiment of the present application.
- FIG. 6 is a schematic flowchart of supplementing parallel entity tuples according to an embodiment of the present application.
- FIG. 7 is a schematic diagram of a process of supplementing subordinate entity tuples according to an embodiment of this application.
- FIG. 8 is a basic structural block diagram of a knowledge graph completion device according to an embodiment of this application.
- FIG. 9 is a block diagram of a basic structure of a computer device according to an embodiment of the present application.
- FIG. 1 is a schematic diagram of the basic process of the knowledge graph completion method of this embodiment.
- a knowledge graph completion method includes the following steps:
- the content on the Internet can be obtained through web crawlers, etc., including encyclopedias, articles or documents, etc., but it is not limited thereto.
- Wikipedia as an example, first obtain the Wikipedia page of the vocabulary through a certain vocabulary, extract the text data as the target content, and link to other vocabulary pages through linking methods such as vocabulary links in the Wikipedia, and at the same time, according to the linking method Difference, determine the relationship between the two words, such as synonym, antonym or subordinate relationship, etc., but not limited to this, organize the obtained page content into text data, repeat the above method, and count the large amount of text obtained Data as knowledge information to be processed.
- Named entities are extracted from the knowledge information to be processed through natural language analysis technology. Divide the knowledge information to be processed into sentences, input different sentences into the natural language analysis model, identify the named entities in the sentence, and mark them, and output the sentences with the marked named entities.
- the natural language analysis model can be Some arbitrary computing models that analyze or recognize natural language.
- the seed pattern is determined in the knowledge information to be processed.
- the determination of the seed pattern can be performed manually or through a neural network model that has been trained to converge to extract the seed pattern.
- the seed pattern contains the annotated sentences.
- the annotated content includes the named entities and associations in the sentence.
- a certain amount of seed patterns of the same category are input into the neural network model as a training sample set, and the convergence is achieved through repeated iterations.
- the neural network model is used to extract named entities and associated relationships from natural language sentences, and to define the pattern of named entities and associated relationships extracted from the sentence as the grammatical model of the seed style of the category.
- a triple of "entity-relationship-entity” is generated as an entity tuple.
- the obtained entities are “cat” and “mammal”, and the relationship is “category”.
- Triads (cats, species, mammals) must be generated.
- the entity tuple to which the entity belongs categorizes and supplements the named entity.
- step S1200 specifically includes the following steps:
- the determination of the seed pattern can be performed manually, or through a neural network model that has been trained to converge to extract the seed pattern.
- the seed style contains the annotated sentences, and the annotated content includes the named entities and associations in the sentences.
- a model of named entities and association relationships is created.
- a certain amount of seed patterns of the same category are input to the neural network model as a training sample set, and a convergent neural network model is trained through iterative iterations to extract named entities from natural language sentences And associations.
- the knowledge information to be processed is divided into sentences, and named entities and associated relationships are extracted from each sentence according to the grammatical pattern determined by the above method.
- the sentenced sentences can be input into a neural network model that has been trained to converge, and the named entities and associated relationships in the sentences can be determined according to the output results of the neural network.
- the method of extracting named entities and association relationships through the grammar model only needs to mark a small part of the knowledge information to be processed, thereby effectively reducing the dependence on the marked samples and saving the labor and time costs of marking.
- step S1400 the following steps are also included:
- the entity tuple After the entity tuple is added to the knowledge graph, find in the knowledge graph whether there is a corresponding entity that is the same as the named entity in the entity tuple. Specifically, after the entity tuple is obtained, the named entity is used as a search condition To search for entities that have a mapping relationship with named entities in the knowledge graph.
- the supplemented entity tuple is linked to the corresponding entity in the knowledge graph as a supplementary relationship of the original entity, forming a relationship network of corresponding entities.
- step S1400 the following steps are also included:
- the entity tuple After the entity tuple is added to the knowledge graph, find in the knowledge graph whether there is a corresponding entity that is the same as the named entity in the entity tuple. Specifically, after the entity tuple is obtained, the named entity is used as a search condition To search for entities that have a mapping relationship with named entities in the knowledge graph.
- the named entity is classified according to preset rules, such as a person's name, place name, or book name, etc., but it is not limited to this.
- the classified entity The method can judge the context of the sentence through the natural language analysis model, and use the subject recognition method to determine the semantics of the named entity in the sentence and classify it. Find similar entities of classified named entities in the knowledge graph, similar entities and named entities are in the same category. Specifically, the way to find similar entities may be to match the similarity between the named entity and the same category of entities in the knowledge graph, and sort according to the similarity, to determine the entity of the same category with the highest similarity as the similar entity of the named entity.
- the entity tuple to which the similar entity belongs is extracted, and the named entity is classified and supplemented according to the matching relationship between the named entity and the similar entity. Specifically, the entity tuples to which the extracted similar entities belong are replaced, and the similar entities are replaced with named entities to generate new entity tuples, and a relationship network of named entities is formed according to the newly generated entity tuples.
- Related content in the knowledge graph is extracted, and the named entity is classified and supplemented according to the matching relationship between the named entity and the similar entity.
- the named entity is "Pastoral Cat”
- the found similar entity is "Pear Cat”
- the entity tuple to which the similar entity belongs is (Pear Cat, Type, Mammal), (Pear Cat, Role, Pet), Replace the "pear cat” in the entity tuple to which the similar entity belongs to "pastoral cat” to generate the entity tuple of (pastoral cat, species, mammal), (pastoral cat, role, pet), and supplement the named entity "pastoral” Cat" relationship network.
- the attribute entity, attribute name and attribute value are extracted from the knowledge information to be processed through a preset second extraction rule to form a triplet as an attribute tuple, for example (Zhang San, height, 175CM).
- the extraction method may be a neural network model that has been trained to converge to extract attribute tuples in a sentence, and the knowledge information to be processed is sentenced and input into the above neural network model, which is determined according to the output result of the neural network Attribute tuples in sentences.
- the neural network model uses a certain number of sentences marked with attribute entities, attribute names, and attribute values as the training sample set, and iterates iteratively until the model converges.
- the attribute tuple is linked to the corresponding entity; if there is no corresponding entity in the knowledge graph that is the same as the attribute entity, the similar entity of the attribute entity is searched for in the knowledge graph according to a preset matching rule, and based on the similar entity
- the attribute tuple to which it belongs categorizes and complements the attribute entities.
- the attribute relationship of the entity can also be ensured.
- the entity attribute and attribute value can be extracted again from the knowledge information, which further improves The relationship network of the entities in the knowledge graph and the data generation.
- step S1400 the following steps are also included:
- the similarity calculation will be performed between named entities extracted from the knowledge information to be processed.
- the similarity calculation can be based on the similarity of the sentence where the named entity is located, or through the natural language analysis method to judge the semantics of the sentence and identify the naming
- the entity acts as a component in the sentence, and the similarity of the parallel relationship of named entities is obtained.
- the parallel relationship similarity calculation can be performed from the entity tuple to which the named entity belongs. For example, in the entity tuple, the named entity with the same association relationship and associated entity has a higher parallel relationship similarity.
- the sentence pattern for extracting the parallel entity may be the grammatical feature extracted from the sentence where the triad of the parallel entity is, for example, the named entity is "dog", the parallel entity with the parallel relationship found is "cat", and the parallel entity
- the extracted entity tuple is (cat, species, mammal), that is, according to this extraction method, (dog, species, mammal) can be extracted as a parallel entity tuple, expressing the cat and dog as a parallel relationship, which are both breastfeeding Subordinate relationship of animals.
- the parallel and subordinate relationship of named entities can be obtained by calculation after supplementing the entity tuples, as the relationship supplement of the original entities, and the integrity of the relationship network between entities in the knowledge graph is improved.
- step S1400 the following steps are also included:
- the knowledge graph is searched for the existing entities associated with the named entity in the entity tuple, and the subordination relationship between the named entity and the existing entity is determined according to the inference rules.
- the named entity is used as a search condition to find an existing entity that is the same as or similar to the named entity in the knowledge graph, and extract the entity tuple to which the existing entity belongs.
- the named entity and the other entity An entity establishes an association relationship.
- the supplementary entity tuple is (Zhang San, colleagues, Li Si), find "Zhang San” or “Li Si” in the knowledge graph, and get the existing entity tuple (Li Si, company, company A), That is, the information of "Zhang San's company is A company” can be obtained by logical reasoning.
- the supplementary entity tuple is (tiger, family, feline), and the existing entity tuple is (cat, family , Carnivora), you can reason about the "Tiger belongs to Carnivora" information.
- the named entities are determined, and a tuple of subordinate entities is generated.
- the existing entity tuple is (cat, family, carnivorous order), and (tiger, order, carnivorous order) can be generated as a subordinate entity tuple according to the inference result.
- the subordinate entity tuple Add the extracted subordinate entity tuple to the knowledge graph, and find the entity with the same named entity in the subordinate entity tuple in the knowledge graph. If there is an entity with the same named entity in the knowledge graph, the subordinate entity tuple Linked to the same entity as the named entity, as a complement to the entity relationship network, if there is no same entity as the named entity in the knowledge graph, the named entity will be used as a new entity in the knowledge graph.
- FIG. 8 is a block diagram of the basic structure of a knowledge graph complementation device for implementation.
- the knowledge graph completion device includes: an acquisition module 2100, an extraction module 2200, a processing module 2300, and an execution module 2400.
- the acquisition module is used to acquire the knowledge information to be processed;
- the extraction module is used to extract the named entity and the association relationship from the knowledge information to be processed according to the preset first extraction rule;
- the processing module is used to extract the named entity And the association relationship to generate entity tuples;
- the execution module is used to supplement the entity tuples into the knowledge graph.
- the knowledge graph completion device further includes: a first processing submodule, a second processing submodule, and a first extraction submodule.
- the first processing sub-module is used to determine a seed pattern from the knowledge information to be processed, wherein the seed pattern includes marked sentences;
- the second processing sub-module is used to generate a corresponding Grammatical pattern;
- the first extraction submodule is used to extract named entities and associated relationships from the knowledge information to be processed according to the grammatical pattern.
- the knowledge graph completion device further includes: a first judgment submodule and a first execution submodule.
- the first judgment sub-module is used to judge whether there is a corresponding entity in the knowledge graph that is the same as the named entity; the first execution sub-module is used if the same correspondence with the named entity already exists in the knowledge graph Entities, the entity tuple is linked to the corresponding entity.
- the knowledge graph completion device further includes: a second judgment submodule, a first search submodule, and a second execution submodule.
- the second judgment sub-module is used to judge whether there is a corresponding entity in the knowledge graph that is the same as the named entity;
- the first search sub-module is used if there is no correspondence in the knowledge graph that is the same as the named entity Entities, the similarity entity of the named entity is searched in the knowledge graph according to a preset matching rule;
- the second execution submodule is used to classify and classify the named entity according to the entity tuple to which the similar entity belongs Supplementary relationship.
- the knowledge graph completion device further includes: a second extraction submodule and a third execution submodule.
- the second extraction sub-module is used to extract attribute tuples from the knowledge information to be processed according to a preset second extraction rule; the third execution sub-module is used to supplement the attribute tuple to the knowledge graph in.
- the knowledge graph completion device further includes: a third processing submodule, a third extraction submodule, and a fourth execution submodule.
- the third processing sub-module is used to calculate the parallel relationship similarity between the named entities in the knowledge information to be processed according to a preset calculation rule; the third extraction sub-module is used to calculate the parallel relationship similarity and all The third extraction rule extracts parallel entity tuples from the knowledge information to be processed; a fourth execution submodule is used to supplement the parallel entity tuples to the knowledge graph.
- the knowledge graph completion device further includes: a fourth processing submodule, a fifth processing submodule, and a fifth execution submodule.
- the fourth processing sub-module is used to determine the subordinate relationship between the named entity and the existing entity in the knowledge graph according to a preset inference rule;
- the fifth processing sub-module is used to generate the subordinate entity according to the subordinate relationship Tuple;
- the fifth execution submodule is used to supplement the subordinate entity tuple to the knowledge graph.
- embodiments of the present application further provide a computer device, including a memory and a processor, and the memory stores computer-readable instructions.
- the processor executes the steps of the graph completion method as described in any one of the above.
- FIG. 9 is a block diagram of the basic structure of the computer device of this embodiment.
- the computer device includes a processor, a non-volatile storage medium, a memory, and a network interface connected through a system bus.
- the non-volatile storage medium of the computer device stores an operating system, a database, and computer-readable instructions.
- the database may store a sequence of control information.
- the processor may implement a A knowledge graph completion method.
- the processor of the computer device is used to provide calculation and control capabilities, and support the operation of the entire computer device.
- the memory of the computer device may store computer readable instructions. When the computer readable instructions are executed by the processor, the processor may cause the processor to perform a knowledge graph completion method.
- the network interface of the computer device is used to connect and communicate with the terminal.
- the structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied.
- the specific computer device may include More or fewer parts than shown in the figure, or some combination of parts, or have a different arrangement of parts.
- the processor is used to execute the specific functions of the acquisition module 2100, the extraction module 2200, the processing module 2300, and the execution module 2400 in FIG. 8.
- the memory stores the program codes and various types of data required to execute the above modules.
- the network interface is used for data transmission to user terminals or servers.
- the memory in this embodiment stores the program codes and data required to execute all submodules in the knowledge graph completion device, and the server can call the server program codes and data to execute the functions of all submodules.
- the present application also provides a storage medium storing computer-readable instructions, which when executed by one or more processors, cause the one or more processors to execute the knowledge graph complement described in any of the foregoing embodiments The steps of the full method, wherein the storage medium may be non-volatile.
- the computer program can be stored in a computer-readable storage medium. When executed, it may include the processes of the foregoing method embodiments.
- the aforementioned storage medium may be a non-volatile storage medium such as a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM), etc.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
一种知识图谱补全方法、装置、计算机设备及存储介质。所述方法包括下述步骤:获取待处理的知识信息(S1100);根据预设的第一抽取规则从所述待处理的知识信息中抽取命名实体和关联关系(S1200);根据所述命名实体和关联关系生成实体元组(S1300);将所述实体元组补充到知识图谱中(S1400)。通过上述方法,获取到的数据源仅需要少部分标注,根据有标注的部分数据生成一个文法样式,再通过该文法样式对未标记的数据进行信息抽取,可以有效减少对数据源中标注的依赖性,减少了为数据源添加标注的人力和时间成本,提高知识图谱补全的效率和精确度。
Description
本申请要求于2019年1月8日提交中国专利局、申请号为201910014362.9,发明名称为“知识图谱补全方法、装置、计算机设备及存储介质”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
本申请涉及信息处理技术领域,特别是涉及一种知识图谱补全方法、装置、计算机设备及存储介质。
知识图谱是知识工程中以图的形式组织的知识集群,其由不同类型的实体作为节点、关系作为连接节点的边所构成的。在知识图谱中,实体指真实世界中的客观物体,或者人类思想中的抽象概念,关系则是描述两个实体之间的实际关系。
在已知的知识图谱中,实体类型有人物、事件、组织机构、地点等,而它们之间的关系类型也十分多样化。不同的实体类型所关注的关系也是不同的。例如,对于人物实体之间,常见关系有亲人及朋友关系;对于人与组织机构之间,常见关系有工作单位、毕业院校等。这些已知的实体间的关系在原始的知识图谱中比较稀疏,而实际上实体间还存在大量的隐含关系。
发明人意识到现有机制中,在构建和补全知识图谱时,会将数据源中所有获取到的实体都进行分析,然后建立所有实体、实体属性之间的关联关系。这样,构建出的知识图谱虽然所能涵盖的范围较广,但对于一批样本,不仅需要知道其中每一个样本数据本身,还需要知道其对应的类别标签,非常依赖含有标签的数据源,标记成本较高。
发明内容
本申请实施例能够提供一种降低对含标注数据源的依赖性的知识图谱补全方法、装置、计算机设备及存储介质。
为解决上述技术问题,本申请实施例提供一种知识图谱补全方法,包括以下步骤:获取待处理的知识信息;根据预设的第一抽取规则从所述待处理的知识信息中抽取命名实体和关联关系;根据所述命名实 体和关联关系生成实体元组;将所述实体元组补充到知识图谱中。
为解决上述技术问题,本申请实施例还提供一种知识图谱补全装置,包括:获取模块,用于获取待处理的知识信息;抽取模块,用于根据预设的第一抽取规则从所述待处理的知识信息中抽取命名实体和关联关系;处理模块,用于根据所述命名实体和关联关系生成实体元组;执行模块,用于将所述实体元组补充到知识图谱中。
为解决上述技术问题,本申请实施例还提供一种计算机设备,包括存储器和处理器,所述存储器中存储有计算机可读指令,所述计算机可读指令被所述处理器执行时,使得所述处理器执行一种知识图谱补全方法,所述知识图谱补全方法包括以下步骤:获取待处理的知识信息;根据预设的第一抽取规则从所述待处理的知识信息中抽取命名实体和关联关系;根据所述命名实体和关联关系生成实体元组;将所述实体元组补充到知识图谱中。
为解决上述技术问题,本申请实施例还提供一种计算机存储介质,其上存储有计算机程序,所述计算机程序被处理器执行时,使得所述处理器执行一种知识图谱补全方法,所述知识图谱补全方法包括以下步骤:获取待处理的知识信息;根据预设的第一抽取规则从所述待处理的知识信息中抽取命名实体和关联关系;根据所述命名实体和关联关系生成实体元组;将所述实体元组补充到知识图谱中。
本申请实施例通过预设的抽取规则,仅需要对待处理的知识信息中小部分内容进行标注即可实现命名实体和关联关系的抽取,并根据抽取得到的命名实体和关联关系形成实体元组对知识图谱进行补充,相较于传统的需要对待处理数据中所有信息都进行标注才可以实现知识图谱补全的方式,可以有效地减少对于数据源中标注的依赖性,节省对于数据源标注的人力和时间成本,提高了知识图谱补全的效率和准确性。
图1为本申请实施例知识图谱补全方法的基本流程示意图;
图2为本申请实施例抽取命名实体和关联关系的流程示意图;
图3为本申请实施例对实体元组进行链接的流程示意图;
图4为本申请实施例对命名实体进行归类并补充关系的流程示意图;
图5为本申请实施例补充属性元组的流程示意图;
图6为本申请实施例补充并列实体元组的流程示意图;
图7为本申请实施例补充从属实体元组的流程示意图;
图8为本申请实施例知识图谱补全装置的基本结构框图;
图9为本申请实施例计算机设备基本结构框图。
体地请参阅图1,图1为本实施例知识图谱补全方法的基本流程示意图。如图1所示,一种知识图谱补全方法,包括以下步骤:
S1100、获取待处理的知识信息;
在一些实施方式中,可以通过网络爬虫等方式获取到互联网上的内容,包括百科、文章或者文献等等,但不限于此。以百科为例,通过一定的词汇首先获取到该词汇的百科页面,提取其中的文本数据作为目标内容,通过百科中的词汇链接等链接方式,链接到其他的词汇页面中,同时,根据链接方式的不同,确定两个词汇之间的关系,例如同义、反义或者上下位关系等,但不限于此,将获取到的页面内容整理形成文本数据,重复上述方法,统计获取到的大量文本数据作为待处理的知识信息。
S1200、根据预设的第一抽取规则从所述待处理的知识信息中抽取命名实体和关联关系;
通过自然语言的解析技术对待处理的知识信息进行命名实体的提取。将待处理的知识信息进行分句,将不同的句子输入到自然语言解析模型中,识别句子中的命名实体,并进行标记,输出得到已标记命名实体的句子,其中,自然语言解析模型可以现有的对自然语言进行分析或识别的任意计算模型。
在待处理的知识信息中确定种子样式,种子样式的确定可以通过人工进行,也可以通过已训练至收敛的用于提取种子样式的神经网络模型进行。种子样式中包含有已标注的句子,标注内容包括句子中的命名实体和关联关系,将一定量的同一类别的种子样式作为训练样本集输入到神经网络模型中,通过反复迭代的方式训练得到收敛的神经网络模型,用于从自然语言句式中提取命名实体和关联关系,定义上述在句子中提取得到命名实体和关联关系的模式为该类别种子样式的文法模式。
将待处理的知识信息的句子根据上述文法模式,识别并抽取命名实体和关联关系,具体地,将上述句子输入到上述用于提取命名实体和关联关系的神经网络模式中,根据神经网络的输出分类结果,得到句子中的命名实体和关联关系。
S1300、根据所述命名实体和关联关系生成实体元组;
根据提取得到的命名实体和关联关系,生成“实体-关系-实体”的三元组,作为实体元组,例如,获取到的实体为“猫”和“哺乳动物”,关系为“种类”,即得生成三元组(猫,种类,哺乳动物)。
S1400、将所述实体元组补充到知识图谱中;
将获取到的实体元组添加到知识图谱中,并判断所述知识图谱中 是否存在与实体元组中命名实体相同的对应实体,若所述知识图谱中已存在与命名实体相同的实体,则将实体元组链接到对应实体;若所述知识图谱中不存在与命名实体相同的对应实体,则根据预设的匹配规则在所述知识图谱中查找所述命名实体的相似实体,并根据相似实体所属的实体元组对所述命名实体进行归类并补充关系。
如图2所示,步骤S1200具体包括以下步骤:
S1210、从所述待处理的知识信息中确定种子样式,其中,所述种子样式包括已标记的语句;
种子样式的确定可以通过人工进行,也可以通过已训练至收敛的用于提取种子样式的神经网络模型进行。种子样式中包含有已标注的语句,标注内容包括语句中的命名实体和关联关系。
S1220、根据所述已标记的语句生成对应的文法模式;
获取到一定量从待处理的知识信息中确定的同一类型的已标记的语句,根据语句的共性确定该类语句的文法模式,其中,文法模式即用于在语句中根据自然语言的表达方法提取出命名实体和关联关系的模式。在一些实施方式中,将一定量的同一类别的种子样式作为训练样本集输入到神经网络模型中,通过反复迭代的方式训练得到收敛的神经网络模型,用于从自然语言句式中提取命名实体和关联关系。
S1230、根据所述文法模式在所述待处理的知识信息中抽取命名实体和关联关系;
将待处理的知识信息进行分句,根据上述方法确定的文法模式,对各个句子进行命名实体和关联关系的提取。在一些实施方式中,可以将分句后的句子输入到已训练至收敛的神经网络模型中,根据神经网络的输出结果确定句子中的命名实体和关联关系。
通过文法模式抽取命名实体和关联关系的方法,只需要在待处理的知识信息中进行一小部分的标记,从而有效降低了对已标记样本的依赖性,节省标记的人力和时间成本。
如图3所示,步骤S1400之后还包括以下步骤:
S1510、判断所述知识图谱中是否存在与所述命名实体相同的对应实体;
将实体元组补充到知识图谱中之后,在知识图谱中查找是否存在与实体元组中的命名实体相同的对应实体,具体地,在获取到实体元组后,将其中的命名实体作为搜索条件,查找知识图谱中搜索与命名实体具有映射关系的实体。
S1520、若所述知识图谱中已存在与所述命名实体相同的对应实体,则将所述实体元组链接到所述对应实体;
当在知识图谱中查找到与命名实体相同的对应实体后,将补充进来的实体元组链接到知识图谱中的对应实体上,作为原有实体的补充 关系,形成对应实体的关系网络。
如图4所示,步骤S1400之后还包括以下步骤:
S1610、判断所述知识图谱中是否存在与所述命名实体相同的对应实体;
将实体元组补充到知识图谱中之后,在知识图谱中查找是否存在与实体元组中的命名实体相同的对应实体,具体地,在获取到实体元组后,将其中的命名实体作为搜索条件,查找知识图谱中搜索与命名实体具有映射关系的实体。
S1620、若所述知识图谱中不存在与所述命名实体相同的对应实体,则根据预设的匹配规则在所述知识图谱中查找所述命名实体的相似实体;
当知识图谱中查找不到与命名实体相同的对应实体时,根据预设的规则对命名实体进行分类,例如人名、地名或者书名等等,但不限于此,在一些实施方式中,分类的方式可以通过自然语言解析模型对句子的语境进行判断,采用主体识别的方式确定命名实体在句子中所占的语义,并进行分类。在知识图谱中查找分类后的命名实体的相似实体,相似实体与命名实体为同一类别。具体地,查找相似实体的方式可以通过匹配命名实体与知识图谱中同一类别实体的相似度,根据相似度进行排序,确定其中相似度最高的同一类别实体作为命名实体的相似实体。
S1630、根据所述相似实体所属的实体元组对所述命名实体进行归类并补充关系;
提取出相似实体所属的实体元组,根据命名实体与相似实体的匹配关系,对命名实体进行归类并补充关系。具体地,将提取出的相似实体所属的实体元组进行替换,将其中的相似实体替换为命名实体,生成新的实体元组,并根据新生成的实体元组形成命名实体的关系网络,补充知识图谱中的相关内容。例如,获取到命名实体为“田园猫”,查找出的相似实体为“梨花猫”,相似实体所属的实体元组有(梨花猫,种类,哺乳动物),(梨花猫,作用,宠物),将相似实体所属的实体元组中的“梨花猫”替换为“田园猫”,生成(田园猫,种类,哺乳动物),(田园猫,作用,宠物)的实体元组,补充命名实体“田园猫”的关系网络。
如图5所示,还包括下述步骤:
S2100、根据预设的第二抽取规则从所述待处理的知识信息中抽取属性元组;
通过预设的第二抽取规则从待处理的知识信息中抽取属性实体、属性名称和属性值,形成三元组,作为属性元组,例如(张三,身高,175CM)。具体地,抽取的方式可以通过已训练至收敛用于在句子中 抽取属性元组的神经网络模型,将待处理的知识信息分句后输入到上述神经网络模型中,根据神经网络的输出结果确定句子中的属性元组,该神经网络模型在训练时,将一定量已标记属性实体、属性名称和属性值的句子作为训练样本集,通过反复迭代的方式直至模型收敛。
S2200、将所述属性元组补充到所述知识图谱中;
将获取到的属性元组添加到知识图谱中,并判断所述知识图谱中是否存在与属性元组中命名实体相同的实体,若所述知识图谱中已存在与属性实体相同的实体,则将属性元组链接到对应实体;若所述知识图谱中不存在与属性实体相同的对应实体,则根据预设的匹配规则在所述知识图谱中查找所述属性实体的相似实体,并根据相似实体所属的属性元组对所述属性实体进行归类并补充关系。
通过上述方法,在实体与关联关系补充的同时,也可以保证实体的属性关系的补充,在抽取命名实体和关联关系后,还可以在知识信息中再次抽取出实体属性和属性值,进一步完善了知识图谱中实体的各项关系和数据生成的关系网络。
如图6所示,步骤S1400之后还包括下述步骤:
S1710、根据预设的计算规则计算所述待处理的知识信息中各命名实体间的并列关系相似度;
将从待处理的知识信息中抽取出的命名实体间进行相似度计算,相似度计算可以依据命名实体所在的句子的相似度,或通过自然语言解析方法对于句子的语义进行判断,并识别出命名实体在句中所充当的成分,得出命名实体的并列关系相似度。另一方面,可以从命名实体所属的实体元组进行并列关系相似度计算,例如实体元组中,关联关系和关联实体相同的命名实体,并列关系相似度较高。
S1720、根据所述并列关系相似度和所述第三抽取规则在所述待处理的知识信息中抽取并列实体元组;
在已有的实体元组中查找出与命名实体并列关系相似度较高的并列实体,根据抽取并列实体的语句模式,在待处理的知识信息中抽取并列实体元组,并确定并列实体的上下位关系。抽取并列实体的语句模式可以是并列实体三元体在所处的句子中抽取出来的语法特征,例如命名实体为“狗”,查找出的具有并列关系的并列实体为“猫”,并且并列实体抽取出的实体元组为(猫,物种,哺乳动物),即根据该抽取方法,可以抽取出(狗,物种,哺乳动物)作为并列实体元组,表达猫和狗作为并列关系,同属于哺乳动物的下位关系。
S1730、将所述并列实体元组补充到所述知识图谱中。
将抽取到的并列实体元组补充到知识图谱中,并在知识图谱中查找并列实体元组中命名实体相同的实体,若知识图谱中存在与上述命名实体相同的实体,则将并列实体元组链接到与命名实体相同的实体 上,作为实体关系网络的补充,若知识图谱中不存在与命名实体相同的实体,则将该命名实体在知识图谱中作为新的实体。
通过上述并列实体元组的补充方法,可以在补充实体元组之后通过计算得到命名实体的并列和上下位关系,作为原有实体的关系补充,提高了知识图谱中实体间关系网络的完整度。
如图7所示,步骤S1400之后还包括下述步骤:
S1810、根据预设的推理规则判断所述命名实体与所述知识图谱中已存在的实体间的从属关系;
在将实体元组补充到知识图谱之后,在知识图谱中查找与实体元组中的命名实体相关联的已有实体,并且根据推理规则判断命名实体与已有实体间的从属关系。具体地,以命名实体作为搜索条件,在知识图谱中查找与命名实体相同或相似的已有实体,并提取出已有实体所属的实体元组,根据实体元组的关系,将命名实体与另一个实体建立关联关系。例如,补充的实体元组为(张三,同事,李四),在知识图谱中查找“张三”或“李四”,得到已有的实体元组(李四,公司,A企业),即可以逻辑推理的方式获得“张三的公司是A企业”的信息,另一方面,补充的实体元组为(老虎,科,猫科),已有的实体元组为(猫科,目,食肉目),可以推理得到“老虎属于食肉目”的信息。
S1820、根据所述从属关系生成从属实体元组;
根据获取到的从属关系,确定其中的命名实体,生成从属实体元组。例如,根据(老虎,科,猫科),已有的实体元组为(猫科,目,食肉目),可以根据推理结果生成(老虎,目,食肉目)作为从属实体元组。
S1830、将所述从属实体元组补充到所述知识图谱中。
将抽取到的从属实体元组补充到知识图谱中,并在知识图谱中查找从属实体元组中命名实体相同的实体,若知识图谱中存在与上述命名实体相同的实体,则将从属实体元组链接到与命名实体相同的实体上,作为实体关系网络的补充,若知识图谱中不存在与命名实体相同的实体,则将该命名实体在知识图谱中作为新的实体。
通过上述从属实体元组的补充方法,可以根据补充的实体元组和已有的实体元组推理得到新的从属关系,作为原有实体的关系补充,提高了知识图谱中实体间关系网络的完整度。
为解决上述技术问题,本申请实施例还提供一种知识图谱补全装置。请参阅图8,图8为本实施知识图谱补全装置的基本结构框图。
如图8所示,知识图谱补全装置,包括:获取模块2100、抽取模块2200、处理模块2300和执行模块2400。其中,获取模块用于获取待处理的知识信息;抽取模块用于根据预设的第一抽取规则从所述待 处理的知识信息中抽取命名实体和关联关系;处理模块用于根据所述命名实体和关联关系生成实体元组;执行模块用于将所述实体元组补充到知识图谱中。
通过预设的抽取规则,仅需要对待处理的知识信息中小部分内容进行标注即可实现命名实体和关联关系的抽取,并根据抽取得到的命名实体和关联关系形成实体元组对知识图谱进行补充,相较于传统的需要对待处理数据中所有信息都进行标注才可以实现知识图谱补全的方式,可以有效地减少对于数据源中标注的依赖性,节省对于数据源标注的人力和时间成本,提高了知识图谱补全的效率和准确性。
在一些实施方式中,知识图谱补全装置还包括:第一处理子模块、第二处理子模块、第一抽取子模块。其中第一处理子模块用于从所述待处理的知识信息中确定种子样式,其中,所述种子样式包括已标记的语句;第二处理子模块用于根据所述已标记的语句生成对应的文法模式;第一抽取子模块用于根据所述文法模式在所述待处理的知识信息中抽取命名实体和关联关系。
在一些实施方式中,知识图谱补全装置还包括:第一判断子模块、第一执行子模块。其中,第一判断子模块用于判断所述知识图谱中是否存在与所述命名实体相同的对应实体;第一执行子模块用于若所述知识图谱中已存在与所述命名实体相同的对应实体,则将所述实体元组链接到所述对应实体。
在一些实施方式中,知识图谱补全装置还包括:第二判断子模块、第一查找子模块、第二执行子模块。其中,第二判断子模块用于判断所述知识图谱中是否存在与所述命名实体相同的对应实体;第一查找子模块用于若所述知识图谱中不存在与所述命名实体相同的对应实体,则根据预设的匹配规则在所述知识图谱中查找所述命名实体的相似实体;第二执行子模块用于根据所述相似实体所属的实体元组对所述命名实体进行归类并补充关系。
在一些实施方式中,知识图谱补全装置还包括:第二抽取子模块、第三执行子模块。其中,第二抽取子模块用于根据预设的第二抽取规则从所述待处理的知识信息中抽取属性元组;第三执行子模块用于将所述属性元组补充到所述知识图谱中。
在一些实施方式中,知识图谱补全装置还包括:第三处理子模块、第三抽取子模块、第四执行子模块。其中,第三处理子模块用于根据预设的计算规则计算所述待处理的知识信息中各命名实体间的并列关系相似度;第三抽取子模块用于根据所述并列关系相似度和所述第三抽取规则在所述待处理的知识信息中抽取并列实体元组;第四执行子模块用于将所述并列实体元组补充到所述知识图谱中。
在一些实施方式中,知识图谱补全装置还包括:第四处理子模块、 第五处理子模块、第五执行子模块。其中,第四处理子模块用于根据预设的推理规则判断所述命名实体与所述知识图谱中已存在的实体间的从属关系;第五处理子模块用于根据所述从属关系生成从属实体元组;第五执行子模块用于将所述从属实体元组补充到所述知识图谱中。
为解决上述技术问题,本申请实施例还提供一种计算机设备,包括存储器和处理器,所述存储器中存储有计算机可读指令,所述计算机可读指令被所述处理器执行时,使得所述处理器执行如上任一项所述的图谱补全方法的步骤。具体请参阅图9,图9为本实施例计算机设备基本结构框图。
如图9所示,计算机设备的内部结构示意图。如图9所示,该计算机设备包括通过系统总线连接的处理器、非易失性存储介质、存储器和网络接口。其中,该计算机设备的非易失性存储介质存储有操作系统、数据库和计算机可读指令,数据库中可存储有控件信息序列,该计算机可读指令被处理器执行时,可使得处理器实现一种知识图谱补全方法。该计算机设备的处理器用于提供计算和控制能力,支撑整个计算机设备的运行。该计算机设备的存储器中可存储有计算机可读指令,该计算机可读指令被处理器执行时,可使得处理器执行一种知识图谱补全方法。该计算机设备的网络接口用于与终端连接通信。本领域技术人员可以理解,图中示出的结构,仅仅是与本申请方案相关的部分结构的框图,并不构成对本申请方案所应用于其上的计算机设备的限定,具体的计算机设备可以包括比图中所示更多或更少的部件,或者组合某些部件,或者具有不同的部件布置。
本实施方式中处理器用于执行图8中获取模块2100、抽取模块2200、处理模块2300和执行模块2400的具体功能,存储器存储有执行上述模块所需的程序代码和各类数据。网络接口用于向用户终端或服务器之间的数据传输。本实施方式中的存储器存储有知识图谱补全装置中执行所有子模块所需的程序代码及数据,服务器能够调用服务器的程序代码及数据执行所有子模块的功能。
本申请还提供一种存储有计算机可读指令的存储介质,所述计算机可读指令被一个或多个处理器执行时,使得一个或多个处理器执行上述任一实施例所述知识图谱补全方法的步骤,其中,存储介质可以为非易失性的。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程,是可以通过计算机程序来指令相关的硬件来完成,该计算机程序可存储于一计算机可读取存储介质中,该程序在执行时,可包括如上述各方法的实施例的流程。其中,前述的存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory,ROM)等非易失性存 储介质,或随机存储记忆体(Random Access Memory,RAM)等。
Claims (20)
- 一种知识图谱补全方法,包括以下步骤:获取待处理的知识信息;根据预设的第一抽取规则从所述待处理的知识信息中抽取命名实体和关联关系;根据所述命名实体和关联关系生成实体元组;将所述实体元组补充到知识图谱中。
- 如权利要求1所述的知识图谱补全方法,所述根据预设的第一抽取规则从所述待处理的知识信息中抽取命名实体和关联关系的步骤,包括以下步骤:从所述待处理的知识信息中确定种子样式,其中,所述种子样式包括已标记的语句;根据所述已标记的语句生成对应的文法模式;根据所述文法模式在所述待处理的知识信息中抽取命名实体和关联关系。
- 如权利要求1所述的知识图谱补全方法,所述将所述实体元组补充到知识图谱中的步骤之后,包括以下步骤:判断所述知识图谱中是否存在与所述命名实体相同的对应实体;若所述知识图谱中已存在与所述命名实体相同的对应实体,则将所述实体元组链接到所述对应实体。
- 如权利要求1所述的知识图谱补全方法,所述将所述实体元组补充到知识图谱中的步骤之后,包括以下步骤:判断所述知识图谱中是否存在与所述命名实体相同的对应实体;若所述知识图谱中不存在与所述命名实体相同的对应实体,则根据预设的匹配规则在所述知识图谱中查找所述命名实体的相似实体;根据所述相似实体所属的实体元组对所述命名实体进行归类并补充关系。
- 如权利要求1所述的知识图谱补全方法,包括下述步骤:根据预设的第二抽取规则从所述待处理的知识信息中抽取属性元组;将所述属性元组补充到所述知识图谱中。
- 如权利要求1-5任一项所述的知识图谱补全方法,所述将所述实体元组补充到知识图谱中的步骤之后,包括下述步骤:根据预设的计算规则计算所述待处理的知识信息中各命名实体间的并列关系相似度;根据所述并列关系相似度和第三抽取规则在所述待处理的知识信息中抽取并列实体元组;将所述并列实体元组补充到所述知识图谱中。
- 如权利要求1-5任一项所述的知识图谱补全方法,所述将所述实体元组补充到知识图谱中的步骤之后,包括下述步骤:根据预设的推理规则判断所述命名实体与所述知识图谱中已存在的实体间的从属关系;根据所述从属关系生成从属实体元组;将所述从属实体元组补充到所述知识图谱中。
- 一种知识图谱补全装置,包括:获取模块,用于获取待处理的知识信息;抽取模块,用于根据预设的第一抽取规则从所述待处理的知识信息中抽取命名实体和关联关系处理模块,用于根据所述命名实体和关联关系生成实体元组;执行模块,用于将所述实体元组补充到知识图谱中。
- 一种计算机设备,包括:处理器;用于存储处理器可执行指令的存储器;其中,所述计算机可执行指令被所述处理器执行时,使得所述处理器执行一种知识图谱补全方法,所述知识图谱补全方法包括以下步骤:获取待处理的知识信息;根据预设的第一抽取规则从所述待处理的知识信息中抽取命名实体和关联关系;根据所述命名实体和关联关系生成实体元组;将所述实体元组补充到知识图谱中。
- 如权利要求9所述的计算机设备,所述根据预设的第一抽取规则从所述待处理的知识信息中抽取命名实体和关联关系的步骤,包括以下步骤:从所述待处理的知识信息中确定种子样式,其中,所述种子样式包括已标记的语句;根据所述已标记的语句生成对应的文法模式;根据所述文法模式在所述待处理的知识信息中抽取命名实体和关联关系。
- 如权利要求9所述的计算机设备,所述将所述实体元组补充到知识图谱中的步骤之后,包括以下步骤:判断所述知识图谱中是否存在与所述命名实体相同的对应实体;若所述知识图谱中已存在与所述命名实体相同的对应实体,则将所述实体元组链接到所述对应实体。
- 如权利要求9所述的计算机设备,所述将所述实体元组补充 到知识图谱中的步骤之后,包括以下步骤:判断所述知识图谱中是否存在与所述命名实体相同的对应实体;若所述知识图谱中不存在与所述命名实体相同的对应实体,则根据预设的匹配规则在所述知识图谱中查找所述命名实体的相似实体;根据所述相似实体所属的实体元组对所述命名实体进行归类并补充关系。
- 如权利要求9所述的计算机设备,包括下述步骤:根据预设的第二抽取规则从所述待处理的知识信息中抽取属性元组;将所述属性元组补充到所述知识图谱中。
- 一种计算机存储介质,其上存储有计算机程序,所述计算机程序被处理器执行时,使得所述处理器执行一种知识图谱补全方法,所述知识图谱补全方法包括以下步骤:获取待处理的知识信息;根据预设的第一抽取规则从所述待处理的知识信息中抽取命名实体和关联关系;根据所述命名实体和关联关系生成实体元组;将所述实体元组补充到知识图谱中。
- 如权利要求14所述的非易失性存储介质,所述根据预设的第一抽取规则从所述待处理的知识信息中抽取命名实体和关联关系的步骤,包括以下步骤:从所述待处理的知识信息中确定种子样式,其中,所述种子样式包括已标记的语句;根据所述已标记的语句生成对应的文法模式;根据所述文法模式在所述待处理的知识信息中抽取命名实体和关联关系。
- 如权利要求14所述的非易失性存储介质,所述将所述实体元组补充到知识图谱中的步骤之后,包括以下步骤:判断所述知识图谱中是否存在与所述命名实体相同的对应实体;若所述知识图谱中已存在与所述命名实体相同的对应实体,则将所述实体元组链接到所述对应实体。
- 如权利要求14所述的非易失性存储介质,所述将所述实体元组补充到知识图谱中的步骤之后,包括以下步骤:判断所述知识图谱中是否存在与所述命名实体相同的对应实体;若所述知识图谱中不存在与所述命名实体相同的对应实体,则根据预设的匹配规则在所述知识图谱中查找所述命名实体的相似实体;根据所述相似实体所属的实体元组对所述命名实体进行归类并补充关系。
- 如权利要求14所述的非易失性存储介质,包括下述步骤:根据预设的第二抽取规则从所述待处理的知识信息中抽取属性元组;将所述属性元组补充到所述知识图谱中。
- 如权利要求14-18任一项所述的非易失性存储介质,所述将所述实体元组补充到知识图谱中的步骤之后,包括下述步骤:根据预设的计算规则计算所述待处理的知识信息中各命名实体间的并列关系相似度;根据所述并列关系相似度和第三抽取规则在所述待处理的知识信息中抽取并列实体元组;将所述并列实体元组补充到所述知识图谱中。
- 如权利要求14-18任一项所述的非易失性存储介质,所述将所述实体元组补充到知识图谱中的步骤之后,包括下述步骤:根据预设的推理规则判断所述命名实体与所述知识图谱中已存在的实体间的从属关系;根据所述从属关系生成从属实体元组;将所述从属实体元组补充到所述知识图谱中。
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910014362.9 | 2019-01-08 | ||
| CN201910014362.9A CN109885691B (zh) | 2019-01-08 | 2019-01-08 | 知识图谱补全方法、装置、计算机设备及存储介质 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2020143319A1 true WO2020143319A1 (zh) | 2020-07-16 |
Family
ID=66925686
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2019/118160 Ceased WO2020143319A1 (zh) | 2019-01-08 | 2019-11-13 | 知识图谱补全方法、装置、计算机设备及存储介质 |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN109885691B (zh) |
| WO (1) | WO2020143319A1 (zh) |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111930962A (zh) * | 2020-09-02 | 2020-11-13 | 平安国际智慧城市科技股份有限公司 | 文献数据价值评估方法、装置、电子设备及存储介质 |
| CN112395425A (zh) * | 2020-10-27 | 2021-02-23 | 腾讯科技(深圳)有限公司 | 一种数据处理方法、装置、计算机设备以及可读存储介质 |
| CN112818132A (zh) * | 2021-02-04 | 2021-05-18 | 上海松鼠课堂人工智能科技有限公司 | 基于认知图谱补全知识点关联性的方法及系统 |
| CN113934864A (zh) * | 2021-10-20 | 2022-01-14 | 中国船舶工业综合技术经济研究院 | 基于知识图谱的军工科研生产数据相似性判别方法及系统 |
| CN114064935A (zh) * | 2021-12-01 | 2022-02-18 | 见知数据科技(上海)有限公司 | 信息图谱构建方法、装置、设备、存储介质和程序产品 |
| CN114117074A (zh) * | 2021-11-30 | 2022-03-01 | 北京百度网讯科技有限公司 | 数据处理方法、装置、电子设备和介质 |
| CN114218950A (zh) * | 2021-12-16 | 2022-03-22 | 中国建设银行股份有限公司 | 异常实体识别方法、装置、设备、存储介质及程序产品 |
| CN114297317A (zh) * | 2021-12-01 | 2022-04-08 | 国能(北京)商务网络有限公司 | 数据处理方法、装置、电子设备及存储介质 |
| CN114331774A (zh) * | 2021-12-31 | 2022-04-12 | 讯飞智元信息科技有限公司 | 学生归寝判断方法、装置、电子设备及存储介质 |
| CN114580411A (zh) * | 2021-12-27 | 2022-06-03 | 天翼云科技有限公司 | 一种关系抽取数据集构建方法、装置及电子设备 |
| CN114780752A (zh) * | 2022-05-12 | 2022-07-22 | 招商银行股份有限公司 | 联邦知识图谱构建方法、系统、设备及存储介质 |
| CN115237382A (zh) * | 2022-07-27 | 2022-10-25 | 中国银行股份有限公司 | 产品关联影响分析方法及装置、存储介质及电子设备 |
| CN115374222A (zh) * | 2021-05-19 | 2022-11-22 | 中移(苏州)软件技术有限公司 | 一种知识图谱构建方法、装置和存储介质 |
| CN115660084A (zh) * | 2022-10-18 | 2023-01-31 | 超聚变数字技术有限公司 | 知识抽取方法、设备及计算机存储介质 |
| CN116127962A (zh) * | 2022-08-02 | 2023-05-16 | 马上消费金融股份有限公司 | 命名实体标注方法、装置、电子设备及存储介质 |
| CN116383312A (zh) * | 2023-01-18 | 2023-07-04 | 合肥中科昂辉科技有限公司 | 一种汽车故障知识图谱构建方法 |
| CN116455426A (zh) * | 2023-04-19 | 2023-07-18 | 广东电网有限责任公司 | 一种电力线通信色噪声分析方法、装置、设备及存储介质 |
| CN118227800A (zh) * | 2024-03-22 | 2024-06-21 | 山东省计算中心(国家超级计算济南中心) | 融合关系描述和关系传递的知识图谱补全方法及系统 |
| CN119003791A (zh) * | 2024-10-18 | 2024-11-22 | 北京北大软件工程股份有限公司 | 篇章级关系抽取方法、装置、设备、介质及产品 |
| US12541692B2 (en) * | 2022-04-29 | 2026-02-03 | Huazhong University Of Science And Technology | Interference-based method for knowledge graph completion and system thereof |
Families Citing this family (28)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109885691B (zh) * | 2019-01-08 | 2024-06-25 | 平安科技(深圳)有限公司 | 知识图谱补全方法、装置、计算机设备及存储介质 |
| CN110489513A (zh) * | 2019-06-24 | 2019-11-22 | 覃立万 | 一种智能机器人社交信息处理方法及与人的社交系统 |
| CN110275966B (zh) * | 2019-07-01 | 2021-10-01 | 科大讯飞(苏州)科技有限公司 | 一种知识抽取方法及装置 |
| CN110489561B (zh) * | 2019-07-12 | 2024-07-12 | 平安科技(深圳)有限公司 | 知识图谱构建方法、装置、计算机设备和存储介质 |
| CN110427524B (zh) * | 2019-08-05 | 2022-03-08 | 北京百度网讯科技有限公司 | 知识图谱补全的方法、装置、电子设备及存储介质 |
| CN110427496B (zh) * | 2019-08-05 | 2022-06-03 | 北京百度网讯科技有限公司 | 用于文本处理的知识图谱扩充方法及装置 |
| CN110569365A (zh) * | 2019-08-29 | 2019-12-13 | 南京智慧光信息科技研究院有限公司 | 基于人工智能的移民关系知识图谱生成方法和机器人系统 |
| CN110781310A (zh) * | 2019-09-09 | 2020-02-11 | 深圳壹账通智能科技有限公司 | 目标概念图谱的构建方法、装置、计算机设备和存储介质 |
| CN111061883B (zh) * | 2019-10-25 | 2023-12-08 | 珠海格力电器股份有限公司 | 更新知识图谱的方法、装置、设备及存储介质 |
| CN111048167B (zh) * | 2019-10-31 | 2023-08-18 | 中电药明数据科技(成都)有限公司 | 一种层级式病例结构化方法及系统 |
| CN112905712B (zh) * | 2019-12-04 | 2023-08-15 | 北京百度网讯科技有限公司 | 知识图谱处理方法、装置、电子设备和介质 |
| CN111368097B (zh) * | 2020-03-30 | 2024-07-30 | 中国建设银行股份有限公司 | 一种知识图谱抽取方法及装置 |
| CN113468335B (zh) * | 2020-03-30 | 2025-04-11 | 海信集团有限公司 | 实体隐含关系抽取方法及设备 |
| CN111666969B (zh) * | 2020-04-22 | 2021-11-23 | 北京百度网讯科技有限公司 | 计算图文相似度的方法、装置、电子设备和可读存储介质 |
| CN111581397A (zh) * | 2020-05-07 | 2020-08-25 | 南方电网科学研究院有限责任公司 | 一种基于知识图谱的网络攻击溯源方法、装置及设备 |
| CN111737369B (zh) * | 2020-05-21 | 2024-07-23 | 北京明略软件系统有限公司 | 一种关系图谱的更新方法、装置及存储介质 |
| CN112100324B (zh) * | 2020-08-28 | 2023-05-05 | 广州探迹科技有限公司 | 一种知识图谱的扩展方法、装置、存储介质和计算设备 |
| CN111930897B (zh) * | 2020-09-14 | 2021-04-27 | 平安国际智慧城市科技股份有限公司 | 专利检索方法、装置、电子设备及计算机可读存储介质 |
| CN112132444B (zh) * | 2020-09-18 | 2023-05-12 | 北京信息科技大学 | 一种互联网+环境下文化创新企业知识缺口的识别方法 |
| CN112486568B (zh) * | 2020-12-02 | 2022-06-28 | 浙江理工大学 | 一种基于知识图谱的程序自动修正方法 |
| CN112463986B (zh) * | 2020-12-08 | 2024-08-06 | 南京明略科技有限公司 | 信息存储的方法及装置 |
| CN112612902B (zh) * | 2020-12-23 | 2023-07-14 | 国网浙江省电力有限公司电力科学研究院 | 一种电网主设备的知识图谱构建方法及设备 |
| CN112597316B (zh) * | 2020-12-30 | 2023-12-26 | 厦门渊亭信息科技有限公司 | 一种可解释性推理问答方法及装置 |
| CN113239208A (zh) * | 2021-05-06 | 2021-08-10 | 广东博维创远科技有限公司 | 一种基于知识图谱的标注训练模型 |
| CN114490884B (zh) * | 2021-12-21 | 2023-06-06 | 北京三快在线科技有限公司 | 确定实体关联关系的方法、装置、电子设备及存储介质 |
| CN114691883B (zh) * | 2022-03-07 | 2024-11-05 | 中国人民解放军国防科技大学 | 一种基于知识图谱的信息检索方法 |
| CN114722147A (zh) * | 2022-03-31 | 2022-07-08 | 长沙博为软件技术股份有限公司 | 一种电子病历中传染病史的质控方法、系统、设备及介质 |
| CN116226461A (zh) * | 2022-12-22 | 2023-06-06 | 中国联合网络通信集团有限公司 | 知识图谱的更新方法、装置及存储介质 |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103488724A (zh) * | 2013-09-16 | 2014-01-01 | 复旦大学 | 一种面向图书的阅读领域知识图谱构建方法 |
| CN104933164A (zh) * | 2015-06-26 | 2015-09-23 | 华南理工大学 | 互联网海量数据中命名实体间关系提取方法及其系统 |
| CN106776711A (zh) * | 2016-11-14 | 2017-05-31 | 浙江大学 | 一种基于深度学习的中文医学知识图谱构建方法 |
| CN107358315A (zh) * | 2017-06-26 | 2017-11-17 | 深圳市金立通信设备有限公司 | 一种信息预测方法及终端 |
| CN109885691A (zh) * | 2019-01-08 | 2019-06-14 | 平安科技(深圳)有限公司 | 知识图谱补全方法、装置、计算机设备及存储介质 |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160041986A1 (en) * | 2014-08-08 | 2016-02-11 | Cuong Duc Nguyen | Smart Search Engine |
| CN105938495A (zh) * | 2016-04-29 | 2016-09-14 | 乐视控股(北京)有限公司 | 实体关系识别方法及装置 |
| CN106951499B (zh) * | 2017-03-16 | 2019-09-20 | 中国人民解放军国防科学技术大学 | 一种基于翻译模型的知识图谱表示方法 |
| CN107368468B (zh) * | 2017-06-06 | 2020-11-24 | 广东广业开元科技有限公司 | 一种运维知识图谱的生成方法及系统 |
| CN108804521B (zh) * | 2018-04-27 | 2021-05-14 | 南京柯基数据科技有限公司 | 一种基于知识图谱的问答方法及农业百科问答系统 |
| CN108875051B (zh) * | 2018-06-28 | 2020-04-28 | 中译语通科技股份有限公司 | 面向海量非结构化文本的知识图谱自动构建方法及系统 |
| CN108932340A (zh) * | 2018-07-13 | 2018-12-04 | 华融融通(北京)科技有限公司 | 一种不良资产经营领域下金融知识图谱的构建方法 |
-
2019
- 2019-01-08 CN CN201910014362.9A patent/CN109885691B/zh active Active
- 2019-11-13 WO PCT/CN2019/118160 patent/WO2020143319A1/zh not_active Ceased
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103488724A (zh) * | 2013-09-16 | 2014-01-01 | 复旦大学 | 一种面向图书的阅读领域知识图谱构建方法 |
| CN104933164A (zh) * | 2015-06-26 | 2015-09-23 | 华南理工大学 | 互联网海量数据中命名实体间关系提取方法及其系统 |
| CN106776711A (zh) * | 2016-11-14 | 2017-05-31 | 浙江大学 | 一种基于深度学习的中文医学知识图谱构建方法 |
| CN107358315A (zh) * | 2017-06-26 | 2017-11-17 | 深圳市金立通信设备有限公司 | 一种信息预测方法及终端 |
| CN109885691A (zh) * | 2019-01-08 | 2019-06-14 | 平安科技(深圳)有限公司 | 知识图谱补全方法、装置、计算机设备及存储介质 |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111930962A (zh) * | 2020-09-02 | 2020-11-13 | 平安国际智慧城市科技股份有限公司 | 文献数据价值评估方法、装置、电子设备及存储介质 |
| CN112395425A (zh) * | 2020-10-27 | 2021-02-23 | 腾讯科技(深圳)有限公司 | 一种数据处理方法、装置、计算机设备以及可读存储介质 |
| CN112818132A (zh) * | 2021-02-04 | 2021-05-18 | 上海松鼠课堂人工智能科技有限公司 | 基于认知图谱补全知识点关联性的方法及系统 |
| CN115374222A (zh) * | 2021-05-19 | 2022-11-22 | 中移(苏州)软件技术有限公司 | 一种知识图谱构建方法、装置和存储介质 |
| CN113934864A (zh) * | 2021-10-20 | 2022-01-14 | 中国船舶工业综合技术经济研究院 | 基于知识图谱的军工科研生产数据相似性判别方法及系统 |
| CN113934864B (zh) * | 2021-10-20 | 2025-06-27 | 中国船舶工业综合技术经济研究院 | 基于知识图谱的军工科研生产数据相似性判别方法及系统 |
| CN114117074A (zh) * | 2021-11-30 | 2022-03-01 | 北京百度网讯科技有限公司 | 数据处理方法、装置、电子设备和介质 |
| CN114064935A (zh) * | 2021-12-01 | 2022-02-18 | 见知数据科技(上海)有限公司 | 信息图谱构建方法、装置、设备、存储介质和程序产品 |
| CN114297317A (zh) * | 2021-12-01 | 2022-04-08 | 国能(北京)商务网络有限公司 | 数据处理方法、装置、电子设备及存储介质 |
| CN114218950A (zh) * | 2021-12-16 | 2022-03-22 | 中国建设银行股份有限公司 | 异常实体识别方法、装置、设备、存储介质及程序产品 |
| CN114580411A (zh) * | 2021-12-27 | 2022-06-03 | 天翼云科技有限公司 | 一种关系抽取数据集构建方法、装置及电子设备 |
| CN114331774A (zh) * | 2021-12-31 | 2022-04-12 | 讯飞智元信息科技有限公司 | 学生归寝判断方法、装置、电子设备及存储介质 |
| US12541692B2 (en) * | 2022-04-29 | 2026-02-03 | Huazhong University Of Science And Technology | Interference-based method for knowledge graph completion and system thereof |
| CN114780752A (zh) * | 2022-05-12 | 2022-07-22 | 招商银行股份有限公司 | 联邦知识图谱构建方法、系统、设备及存储介质 |
| CN115237382A (zh) * | 2022-07-27 | 2022-10-25 | 中国银行股份有限公司 | 产品关联影响分析方法及装置、存储介质及电子设备 |
| CN116127962A (zh) * | 2022-08-02 | 2023-05-16 | 马上消费金融股份有限公司 | 命名实体标注方法、装置、电子设备及存储介质 |
| CN115660084A (zh) * | 2022-10-18 | 2023-01-31 | 超聚变数字技术有限公司 | 知识抽取方法、设备及计算机存储介质 |
| CN116383312A (zh) * | 2023-01-18 | 2023-07-04 | 合肥中科昂辉科技有限公司 | 一种汽车故障知识图谱构建方法 |
| CN116455426A (zh) * | 2023-04-19 | 2023-07-18 | 广东电网有限责任公司 | 一种电力线通信色噪声分析方法、装置、设备及存储介质 |
| CN118227800A (zh) * | 2024-03-22 | 2024-06-21 | 山东省计算中心(国家超级计算济南中心) | 融合关系描述和关系传递的知识图谱补全方法及系统 |
| CN119003791A (zh) * | 2024-10-18 | 2024-11-22 | 北京北大软件工程股份有限公司 | 篇章级关系抽取方法、装置、设备、介质及产品 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109885691B (zh) | 2024-06-25 |
| CN109885691A (zh) | 2019-06-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2020143319A1 (zh) | 知识图谱补全方法、装置、计算机设备及存储介质 | |
| CN111753099B (zh) | 一种基于知识图谱增强档案实体关联度的方法及系统 | |
| CN112035653B (zh) | 一种政策关键信息提取方法和装置、存储介质、电子设备 | |
| CN106570148B (zh) | 一种基于卷积神经网络的属性抽取方法 | |
| US11526804B2 (en) | Machine learning model training for reviewing documents | |
| US11556711B2 (en) | Analyzing documents using machine learning | |
| TWI662425B (zh) | 一種自動生成語義相近句子樣本的方法 | |
| WO2017166912A1 (zh) | 商品短文本核心词提取方法和装置 | |
| CN114647713A (zh) | 基于虚拟对抗的知识图谱问答方法、设备及存储介质 | |
| CN113157930A (zh) | 基于多源异构数据的知识图谱构建方法、系统以及终端 | |
| JPWO2014033799A1 (ja) | 単語意味関係抽出装置 | |
| CN101464897A (zh) | 一种词匹配及信息查询方法及装置 | |
| CN117112595B (zh) | 一种信息查询方法、装置、电子设备及存储介质 | |
| CN114840685A (zh) | 一种应急预案知识图谱构建方法 | |
| CN109376261B (zh) | 基于中级文本语义增强空间的模态独立检索方法和系统 | |
| CN106776532A (zh) | 一种知识问答方法及装置 | |
| CN103678287B (zh) | 一种关键词翻译统一的方法 | |
| CN111428047A (zh) | 一种基于ucl语义标引的知识图谱构建方法及装置 | |
| CN112883165B (zh) | 一种基于语义理解的智能全文检索方法及系统 | |
| US20210064700A1 (en) | Removing outliers from training data for machine learning | |
| CN115470358A (zh) | 一种跨语言实体链接方法、系统、设备及终端 | |
| CN110888983A (zh) | 一种正负面情感分析方法、终端设备及存储介质 | |
| CN118626631A (zh) | 基于大语言模型的专利文本推荐方法、系统、设备及介质 | |
| TW201905734A (zh) | 語意分析裝置、方法及其電腦程式產品 | |
| CN113868312A (zh) | 多方法融合的机构匹配方法、装置、设备和存储介质 |
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: 19909203 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: 19909203 Country of ref document: EP Kind code of ref document: A1 |