CN112860916A - Movie-television-oriented multi-level knowledge map generation method - Google Patents

Abstract

The present disclosure provides a film and television-oriented multi-level knowledge graph generation method, which acquires different levels of information data related to film and television to be processed; performs relation extraction on the acquired information data to obtain triplet data, and constructs multi-level data according to the triplet data. A single-level knowledge graph; use the relation triplet data and attribute triplet data in triplet data to embed structure and attributes; combine the results of structure embedding and attribute embedding to perform entity alignment, and align multiple entities after entity alignment. The single-level knowledge graph is integrated to obtain a multi-level knowledge graph.

Description

A multi-level knowledge graph generation method for film and television

technical field

The present disclosure relates to the technical fields of data mining and intelligent information processing, and in particular, to a method for generating a multi-level knowledge graph for film and television.

Background technique

The statements in this section merely provide background related to the present disclosure and do not necessarily constitute prior art.

In the field of film and television, there are problems of many data sources, massive quantities, diverse data forms, and complex data structures. Different knowledge graphs of film and television store different knowledge. These knowledge have many repetitions and can also complement each other. Therefore, some researchers have proposed that they can Integrate various knowledge graphs to form a multi-level knowledge graph. In order to form a multi-level knowledge graph, a basic problem is to align the entity knowledge that exists in different film and television knowledge graphs but represents the same meaning.

Alignment methods are mainly divided into two parts: traditional alignment methods and embedding-based alignment methods. The former mainly uses a supervised machine learning model to align entities through attribute similarity matching. The latter is mainly based on the method of representation learning. Specifically, by mapping the entities and relationships of the film and television knowledge graph to a low-dimensional vector space, and then calculating the similarity between entities, calculation and reasoning are performed. Most of these methods only focus on how to encode relation triples in a better way, while ignoring those attribute triples; especially for entities lacking relations, if only use relation triples to align entities, the resulting The comprehensiveness and accuracy of knowledge graphs are poor.

SUMMARY OF THE INVENTION

In order to solve the deficiencies of the prior art, the present disclosure provides a method for generating a multi-level knowledge graph for film and television. Aligned from the perspective of , and finally formed a multi-level knowledge graph.

In order to achieve the above object, the present disclosure adopts the following technical solutions:

A first aspect of the present disclosure provides a method for generating a multi-level knowledge graph for film and television.

A method for generating a multi-level knowledge graph for film and television, comprising the following steps:

Obtain information data at different levels related to the film and television to be processed;

Relation extraction is performed on the acquired information data to obtain triple data, and multiple single-level knowledge graphs are constructed according to the triple data;

Use relation triple data and attribute triple data in triple data to embed structure and attributes;

Combine the results of structure embedding and attribute embedding for entity alignment, and integrate multiple single-level knowledge graphs after entity alignment to obtain multi-level knowledge graphs.

A second aspect of the present disclosure provides a film and television-oriented multi-level knowledge graph generation system.

A multi-level knowledge graph generation system for film and television: including:

The data acquisition module is configured to: acquire information data at different levels related to the film and television to be processed;

The relationship extraction module is configured to: perform relationship extraction on the acquired information data to obtain triplet data, and construct multiple single-level knowledge graphs according to the triplet data;

an embedding module, configured to: perform structure and attribute embedding using relation triplet data and attribute triplet data in triplet data;

The knowledge graph integration module is configured to: combine the results of structure embedding and attribute embedding to perform entity alignment, and integrate multiple single-level knowledge graphs after entity alignment to obtain a multi-level knowledge graph.

A third aspect of the present disclosure provides a video query method based on a multi-level knowledge graph.

A film and television query method based on a multi-level knowledge graph, comprising the following steps:

Get the text to be queried;

Parse the query text to obtain the parsing result;

According to the analysis result and the multi-level knowledge graph constructed by using the generation method described in the first aspect of the present disclosure, the movie information data query is performed to obtain the query result.

Further, the movie query result is displayed.

A fourth aspect of the present disclosure provides a computer-readable storage medium on which a program is stored, and when the program is executed by a processor, implements the steps in the method for generating a video-oriented multi-level knowledge graph according to the first aspect of the present disclosure .

A fifth aspect of the present disclosure provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, when the processor executes the program, the implementation is as described in the first aspect of the present disclosure The steps in the method for generating a multi-level knowledge graph for film and television.

Compared with the prior art, the beneficial effects of the present disclosure are:

1. The generation method, system, medium or electronic device described in the present disclosure simultaneously aligns the relationship triples and attribute triples existing in the single-level movie knowledge graph from the perspectives of structure and attributes, and finally The formation of a multi-level knowledge map overcomes the structural defects and relationship defects of the previous multi-level knowledge map, and ensures the comprehensiveness and accuracy of the generated multi-level knowledge map.

2. The film and television query method based on the multi-level knowledge graph described in the present disclosure, combined with the constructed comprehensive and accurate multi-level knowledge graph, can realize fast query of relevant film and television information data, and improve the accuracy of the query.

Description of drawings

The accompanying drawings that constitute a part of the present disclosure are used to provide further understanding of the present disclosure, and the exemplary embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation of the present disclosure.

FIG. 1 is a schematic flowchart of a method for generating a video-oriented multi-level knowledge graph according to Embodiment 1 of the present disclosure.

FIG. 2 is a schematic diagram of a multi-level knowledge graph structure provided by Embodiment 1 of the present disclosure.

FIG. 3 is a schematic diagram of an attribute value and attribute type embedding process based on the Pseudo-Siamese Neural Network provided in Embodiment 1 of the present disclosure.

Detailed ways

The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

It should be noted that the following detailed description is exemplary and intended to provide further explanation of the present disclosure. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs.

It should be noted that the terminology used herein is for the purpose of describing specific embodiments only, and is not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly dictates otherwise, the singular is intended to include the plural as well, furthermore, it is to be understood that when the terms "comprising" and/or "including" are used in this specification, it indicates that There are features, steps, operations, devices, components and/or combinations thereof.

The embodiments of this disclosure and features of the embodiments may be combined with each other without conflict.

Example 1:

As shown in FIG. 1 , Embodiment 1 of the present disclosure provides a method for generating a multi-level knowledge graph for film and television, including the following steps:

Obtain information data at different levels related to the film and television to be processed;

Relation extraction is performed on the acquired information data to obtain triple data, and multiple single-level knowledge graphs are constructed according to the triple data;

Use relation triple data and attribute triple data in triple data to embed structure and attributes;

Combine the results of structure embedding and attribute embedding for entity alignment, and integrate multiple single-level knowledge graphs after entity alignment to obtain multi-level knowledge graphs.

Specifically, it includes the following:

S1: First build a single-level knowledge graph

S1.1: First, analyze the different levels of film knowledge. For example, film types include suspense films, horror films, romance films, martial arts films, literary films, etc.; film genres include surreal film genres, minimalist genres, etc.; Technically, it is divided into black and white movies, ordinary color movies, 3D movies, etc. The movie actors also come from different countries and regions in the world, and data is obtained at different levels at this stage.

S1.2: Perform relation extraction according to the different characteristics of the acquired data. Here we only need to acquire three types of data: text, data, and tables. The text uses dependency parsing and HanLP to extract triples, and the data and tables use the Pearson correlation coefficient method to find medium and high correlations between entities to extract triples. The triplet form is <h,r,t> where h is the head entity, t is the tail entity, and r is the relationship between the two entities.

S1.3: Construct a single-level knowledge graph from the extracted triples.

S2.1: Multi-source data fusion

Knowledge is constantly being updated. In order to judge whether the newly added knowledge is authentic, we need to use the multi-source data fusion model to fuse the collected information. Only when the data is judged to be authentic and credible can it be added to the knowledge graph middle.

The working process of the multi-source data fusion model is as follows:

(1) Data segmentation: In the relation extraction stage, knowledge extraction is carried out centered on entity keywords. Therefore, data from different sources are aggregated in blocks based on entity keywords at various levels as candidate matching knowledge. In this way, when data is merged, it can be merged with the data at the respective levels first, avoiding traversing the entire knowledge base and reducing the computational complexity.

(2) Multi-source data fusion coefficient K: The candidate matching knowledge in the same block is used to match the knowledge of the original knowledge base with the multi-source data fusion coefficient K. If K is greater than the set threshold, it is considered that the candidate matching knowledge For correct knowledge, it can be added to the knowledge base.

The multi-source data fusion coefficient K is defined as follows:

K consists of two parts, where confidence is the confidence score, and the latter part is the average of entity similarity and relation similarity. The confidence consists of two parts, M and cf, where M is the confidence of the data source. For example, authoritative websites or knowledge bases such as Baidu Encyclopedia and HowNet have a higher M value. cf is a confidence level calculated for each combination of two entities based on the distance between entity and entity, entity and relation representation.

Entity_sim is the text similarity calculation between entities, Relationship_sim is the relationship similarity calculation, and the average of the two is taken as the similarity of the knowledge. If the corresponding similarity is greater than the set threshold of 0.5, the knowledge is considered more credible. In the formula L represents entity location and R represents relation location. _{Li -L j represents} the distance between entity 1 and entity 2; _Li -R represents the distance between entity 1 and the relationship, the larger the distance, the less likely there is a semantic relationship between the entity and the relationship. , the confidence will be lower.

S3: Building a multi-level knowledge graph

After the multi-source data fusion is completed, the knowledge graph is still a decentralized single-level knowledge graph. Through entity alignment, a single-level knowledge map is integrated to build a multi-level knowledge map. Knowledge exists in the form of triples, which are usually divided into relation triples and attribute triples, such as relation triples (Zhang Ziyi, Husband, Wang Feng), attribute triples (Zhang Ziyi, age, 41 ).

By using relation triples and attribute triples for structure and attribute embedding, entity semantic information and attribute information can be fully utilized, and the accuracy of the alignment process can be further improved. The specific methods are as follows:

S3.1: Use the unified naming method to merge the triples in the single-level knowledge graph through the predicate similarity, and set 0.95 as the similarity threshold, so that entities and relations can be embedded in the same vector space.

Because the representation of the same entity in different knowledge maps is different, it is a simple name unification for the same entity in different knowledge maps. After unification, the subsequent alignment is performed in the vector space.

S3.2: Use relation triples and attribute triples for structure and attribute alignment, respectively.

(A) Structural Embedding: Performed through predicate-aligned triples, using relation triples and training set for structural embedding, learning vector representations of entities and relationships. Given a relation triplet Tt=(h,r,t), expect h+r=t, in order to measure rationality, this model optimizes the marginal-based ranking loss, so that the score of positive triples is lower than that of negative triples The score:

Tr是得分函数，表示所有的正三元组集合，Tr′表示通过随机实体(但不是同时用这两个实体)替换其头部或尾部而生成的相关负三元组集合，α是一个加权正、负三元组的比率超参数，其取值范围为[0,1]。因此我们可以学习实体在KGs(Knowledge Graphs)上的近似向量表示，经过结构嵌入后的实体相似性度量如公式所示：in,

Tr is the score function representing the set of all positive triplets, Tr′ represents the set of correlated negative triplets generated by replacing their heads or tails with random entities (but not both), and α is a weighted positive , the ratio hyperparameter of negative triples, whose value range is [0,1]. Therefore, we can learn the approximate vector representation of entities on KGs (Knowledge Graphs). The entity similarity measure after structural embedding is shown in the formula:

Among them, G1 and G2 refer to knowledge graph 1 and knowledge graph 2, hi, hj refer to the entities in knowledge graph 1 and knowledge graph 2, the cos function refers to the cosine function, and the Sim function refers to The similarity value after structural embedding is equal to the cosine function value of the two entities in the vector space.

(B) Attribute value and attribute type embedding based on Pseudo-Siamese Neural Network

Attribute Value Embedding: A Bidirectional Gated Recurrent Unit (Bi-GRU) network is used to encode attribute values into a single embedding from both directions, as detailed in the following formula:

Z _i =σ(W _Z [C _i ,hi _-1 ])

R _i =σ(W _r [C _i ,hi _-1 ])

Where Z and R are the update gate and reset gate of the GRU unit respectively, W _Z , W _h , W _r are the weight matrix, Bi-GRU is composed of forward GRU and backward GRU, and the forward GRU is read from left to right Embedding of input characters, backward GRU reads the embedding of characters in reverse, if read attribute value v=(c ₀ , c ₁ ..... c _n ), their outputs are respectively

The initial state of Bi-GRU is set to 0 vector, after reading character embedding, attribute value embedding is finally formed

是实体e的m个属性值，计算注意力权重

属性值的权重应该与其属性类型的权重一致，如下公式：Attribute Type Embedding: To learn the importance of different attributes to entities, let attribute types and attribute values share an attention weight. Assumption

are the m attribute values of entity e, and calculate the attention weight

The weight of the attribute value should be consistent with the weight of its attribute type, as follows:

Concatenating the attribute type embedding with the attribute value embedding gives the final entity attribute embedding:

The attribute embedding process is to use the attributes of the entities to be aligned which usually have high similarity, learn on the attribute information of the two KGs through the Pseudo-Siamese Neural Network, and then use some trained metrics to evaluate the similarity between entity embeddings, entities Embedding is ultimately the connection of entity attribute type embedding and attribute value embedding.

S3: Integrate multi-level knowledge graphs:

Finally, the aligned single-level knowledge graph is integrated into a unified multi-level knowledge graph.

Through the process of attribute embedding and structure embedding, the similarity matrix of each entity pair is obtained. Then, the optimal bipartite graph matching algorithm is used as the new entity pair for the next iteration, and finally the optimal matching entity is found, that is, the realization entity Align.

Example 2:

Embodiment 2 of the present disclosure provides a film and television-oriented multi-level knowledge graph generation system, including:

The data acquisition module is configured to: acquire information data at different levels related to the film and television to be processed;

The relationship extraction module is configured to: perform relationship extraction on the acquired information data to obtain triplet data, and construct multiple single-level knowledge graphs according to the triplet data;

an embedding module, configured to: perform structure and attribute embedding using relation triplet data and attribute triplet data in triplet data;

The knowledge graph integration module is configured to: combine the results of structure embedding and attribute embedding to perform entity alignment, and integrate multiple single-level knowledge graphs after entity alignment to obtain a multi-level knowledge graph.

The working method of the system is the same as the method for generating a video-oriented multi-level knowledge graph provided in Embodiment 1, and details are not repeated here.

Example 3:

At present, people mainly search for film and television related knowledge through search engines, which collect web page information from the Internet according to the keywords input by the user, organize and process the information, and display the retrieved relevant web pages to the user. The search engine is suitable for a full range of information retrieval, not limited to a specific field of query, but it has the following disadvantages: First, it often mechanically matches web page information through keywords, and cannot understand the semantics of the sentences entered by the user, so the user input There are certain requirements. The more accurate the keywords entered by the user, the closer the content returned by the search is to the content that the user wants. Second, the search returns relevant web pages, not exact answers. Users have to browse many web pages to get the required information.

In view of this, Embodiment 3 of the present disclosure provides a video query method based on a multi-level knowledge graph, including the following steps:

Get the text to be queried;

Parse the query text to obtain the parsing result;

Perform movie information data query according to the analysis result and the multi-level knowledge graph constructed by the generation method described in Embodiment 1, and obtain the query result;

Display the movie query results.

Among them, the data source of the multi-level knowledge graph is obtained by crawling the movie text data on the web page.

The parsing method is:

The query text is parsed by an intent recognition model based on an enhanced convolutional neural network, and the parsing result is obtained, specifically:

Construct a dictionary tree from entities, attributes and collected professional words for film and television in the knowledge graph;

Through the intent recognition model, the words in the query text are matched with the film and television professional words in the dictionary tree, and the matched movie professional words are used as the word information sequence;

Integrate all the matching film and television professional terms to form the analysis result.

Example 4:

Embodiment 4 of the present disclosure provides a computer-readable storage medium on which a program is stored, and when the program is executed by a processor, implements the steps in the method for generating a video-oriented multi-level knowledge graph according to Embodiment 1 of the present disclosure .

Example 5:

Embodiment 5 of the present disclosure provides an electronic device, including a memory, a processor, and a program stored in the memory and running on the processor. When the processor executes the program, the implementation is as described in Embodiment 1 of the present disclosure. The steps in the method for generating a multi-level knowledge graph for film and television.

As will be appreciated by one skilled in the art, embodiments of the present disclosure may be provided as a method, system, or computer program product. Accordingly, the present disclosure may take the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present disclosure may take the form of a computer program product embodied on one or more computer-usable storage media having computer-usable program code embodied therein, including but not limited to disk storage, optical storage, and the like.

The present disclosure is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the disclosure. It will be understood that each flow and/or block in the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing device to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing device produce Means for implementing the functions specified in a flow or flow of a flowchart and/or a block or blocks of a block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory result in an article of manufacture comprising instruction means, the instructions The apparatus implements the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams.

These computer program instructions can also be loaded on a computer or other programmable data processing device to cause a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process such that The instructions provide steps for implementing the functions specified in the flow or blocks of the flowcharts and/or the block or blocks of the block diagrams.

Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. During execution, the processes of the embodiments of the above-mentioned methods may be included. The storage medium may be a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM) or the like.

The above descriptions are only preferred embodiments of the present disclosure, and are not intended to limit the present disclosure. For those skilled in the art, the present disclosure may have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included within the protection scope of the present disclosure.

Claims (10)

1. A movie-oriented multi-level knowledge map generation method is characterized by comprising the following steps: the method comprises the following steps:

acquiring information data of different layers related to a movie to be processed;

extracting the relation of the acquired information data to obtain triple data, and constructing a plurality of single-level knowledge maps according to the triple data;

embedding the structure and the attribute by utilizing the relation ternary group data and the attribute ternary group data in the ternary group data;

and combining the results of structure embedding and attribute embedding to carry out entity alignment, and integrating a plurality of single-level knowledge maps after entity alignment to obtain a multi-level knowledge map.

2. The method for generating a multi-level knowledge-graph for movies according to claim 1, characterized in that:

the information data includes text, data and tables, the text uses dependency syntax analysis and HanLP to extract triples, and the data and tables extract triples by Pearson's correlation coefficient method.

3. The method for generating a multi-level knowledge-graph for movies according to claim 1, characterized in that:

the method for carrying out multivariate data fusion on the acquired information data and judging the authenticity of the data comprises the following steps:

carrying out block aggregation on data from different sources by taking the entity keywords of each layer as a basis to serve as candidate matching knowledge;

and matching the candidate matching knowledge in the same block by using the multisource data fusion coefficient and the knowledge of the original knowledge base, and if the multisource data fusion coefficient is greater than a set threshold value, considering that the candidate matching knowledge is correct knowledge and can be added into the knowledge base, otherwise, the candidate matching knowledge cannot be added.

4. The method for generating a multi-level knowledge-graph for movies according to claim 1, characterized in that:

merging the triples in the single-level knowledge graph by adopting a uniform naming method through predicate similarity, and embedding the entities and the relations into the same vector space;

in the vector space, structure and attribute alignment is carried out by utilizing the relation triples and the attribute triples respectively;

and integrating the aligned single-level knowledge maps into a uniform multi-level knowledge map.

5. The method for generating a multi-level knowledge-graph for movies according to claim 1, characterized in that:

carrying out structure embedding based on the TransE relation triple, comprising the following steps: and combining the predicate-aligned triples, performing structure embedding by using the relation triples and the training set, learning vector representation of the entities and the relation, and making the scores of the positive triples lower than those of the negative triples to obtain an entity similarity measurement expression after structure embedding.

6. The method for generating a multi-level knowledge-graph for movies according to claim 1, characterized in that:

embedding attribute values and attribute types based on a pseudo-twin neural network, comprising:

and encoding the attribute values into single embedding from two directions by using a bidirectional gating recursive unit network, enabling the attribute types and the attribute values to share an attention weight, and connecting the attribute type embedding and the attribute value embedding in series to obtain a final attribute embedding result.

7. A multi-level knowledge map generation system facing to film and television is characterized in that: the method comprises the following steps:

a data acquisition module configured to: acquiring information data of different layers related to a movie to be processed;

a relationship extraction module configured to: extracting the relation of the acquired information data to obtain triple data, and constructing a plurality of single-level knowledge maps according to the triple data;

an embedding module configured to: embedding the structure and the attribute by utilizing the relation ternary group data and the attribute ternary group data in the ternary group data;

a knowledge-graph integration module configured to: and combining the results of structure embedding and attribute embedding to carry out entity alignment, and integrating a plurality of single-level knowledge maps after entity alignment to obtain a multi-level knowledge map.

8. A movie and television query method based on a multi-level knowledge graph is characterized by comprising the following steps: the method comprises the following steps:

acquiring a text to be queried;

analyzing the text to be queried to obtain an analysis result;

and inquiring the movie information data according to the analysis result and the multi-level knowledge graph constructed by the generation method of any one of claims 1 to 6 to obtain an inquiry result.

9. A computer-readable storage medium, on which a program is stored, wherein the program, when executed by a processor, implements the steps in the movie-oriented multi-level knowledge-graph generation method according to any one of claims 1 to 6.

10. An electronic device comprising a memory, a processor and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method for generating a multi-level knowledge map for movie & TV according to any one of claims 1-6.

Images