CN112364171A - Novel knowledge graph entity portrait method - Google Patents

Novel knowledge graph entity portrait method Download PDF

Info

Publication number
CN112364171A
CN112364171A CN202010954910.9A CN202010954910A CN112364171A CN 112364171 A CN112364171 A CN 112364171A CN 202010954910 A CN202010954910 A CN 202010954910A CN 112364171 A CN112364171 A CN 112364171A
Authority
CN
China
Prior art keywords
entity
label
tags
tag
candidate
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.)
Pending
Application number
CN202010954910.9A
Other languages
Chinese (zh)
Inventor
张祥
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Southeast University
Original Assignee
Southeast University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Southeast University filed Critical Southeast University
Priority to CN202010954910.9A priority Critical patent/CN112364171A/en
Publication of CN112364171A publication Critical patent/CN112364171A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
    • G06F16/367—Ontology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/335—Filtering based on additional data, e.g. user or group profiles
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/34—Browsing; Visualisation therefor

Abstract

The invention provides a novel knowledge graph entity portrait method, which comprises the following steps: step 1, constructing a candidate label pool; step 2, filtering the candidate tags; step 3, evaluating the distinguishability of the candidate labels; step 4, reordering; and 5, visually generating a solid portrait result. The invention focuses on finding the label with distinctiveness, fully embodies the distinctiveness among the entities; the entity tag distinguishing performance is deeply researched through an HAS model and a reordering method, the redundancy in a tag space is reduced, the entity similarity measurement is more comprehensive, and the tag distinguishing capacity is improved. The invention provides the research of the entity portrait in the knowledge map for the first time, which can be a good supplement to the research of the entity abstract and is beneficial to the realization of downstream tasks such as entity linkage, entity recommendation and the like.

Description

Novel knowledge graph entity portrait method
Technical Field
The invention belongs to the technical field of computers, and relates to a novel knowledge graph entity portrayal method.
Background
In recent years, knowledge graph has been developed rapidly, and it stores fact information in the form of "entity-relationship-entity" and "entity-attribute value", and is widely used in tasks such as entity search, data integration, and the like. There are two types of methods for understanding an entity: one is to identify an entity as a corresponding real-world object; another type of approach is to compare one entity with other entities to see its uniqueness. However, the capacity and structural complexity of the knowledge-graph greatly reduces the efficiency of identifying and comparing entities. In order to solve the problem of entity identification in entity understanding, the field of entity summarization has gained wide attention in recent years. The method for entity abstract shortens the lengthy entity description by extracting a concise abstract and retains important information content in the abstract.
Although the entity abstract can help a user to quickly understand the entity, the user still has difficulty in understanding the entity by only relying on the abstract, and the problem of distinguishing the entities is still not solved. Because the entity abstract only contains local information of the entity, but lacks global information reflecting the uniqueness of the entity relative to other entities, the distinguishability of the entity cannot be shown in the abstract.
Disclosure of Invention
In order to solve the problems, the invention provides an entity portrayal method based on a knowledge graph, and provides an extensible representation learning model, namely an HAS model, which is used for generating multi-mode entity embedding and can efficiently find out entity labels with the most distinctiveness in the knowledge graph, thereby embodying the unique characteristics of the entities in the knowledge graph.
In order to achieve the purpose, the invention provides the following technical scheme:
a novel knowledge graph entity portrait method comprises the following steps:
step 1: building a pool of candidate tags
Giving a knowledge graph as input, automatically enumerating all possible labels for each entity type, and putting the labels into a label pool;
step 2: filtering candidate tags
Filtering the low-quality labels through a filter to obtain a candidate label set;
and step 3: evaluating candidate tag distinctiveness
Measuring the distinguishability of the label through a uniqueness evaluator, calculating the similarity between positive examples of entities matched with the label and between positive examples and negative examples of entities not matched with the label, and evaluating the distinguishability of the label;
and 4, step 4: reordering
Selecting a label with distinctiveness by a reordering method to generate a final label set;
and 5: visual generation of entity portrait results
And traversing all entity label descriptions, searching whether a certain entity is matched with certain labels or not, obtaining a final portrait result, and visually presenting the portrait result to a user.
Further, in step 1, the tags include attribute tags and relationship tags, where the attribute tags include attribute interval tags and attribute value tags, and the relationship tags include relationship-entity tags and relationship-attribute tags; the process of generating the label is as follows: tags are automatically generated by violently enumerating all attributes and combinations of attribute values, or combinations of relationships and entities, from the knowledge-graph.
Furthermore, when the attribute interval label is generated, the continuous value of the attribute is discretized, and a proper interval is divided by adopting a method of combining equal-width discretization and density discretization.
Further, in the step 2, heuristic rules are used to filter the low-quality labels, including the labels without representativeness and the labels without distinctiveness.
Further, the heuristic rule specifically includes:
for an entity type t and candidate labels l associated with t, ε istIs defined as the set of all entities under type t, and
Figure BDA0002678257290000021
Figure BDA0002678257290000022
represents a set of proper instances of the entity associated with tag l, the ratio support (l) of the proper instances when matching the tag l satisfies
Figure BDA0002678257290000023
Then, the label is considered as a high-quality candidate label, and alpha and 1-alpha respectively represent an upper bound and a lower bound; when support (l) is out of the upper and lower bounds, it is considered a low quality tag.
Further, in the step 3, a multi-mode entity representation model is adopted to measure the similarity between entities, and meanwhile, a homogeneity mode, an attribute equivalence mode and a structure equivalence mode are considered to generate 3 random walk paths; then, mixing the 3 paths according to a certain proportion, and following the skip-gram learning process of deep walk to obtain the embedded representation of the entity; finally, the cosine similarity is used for similarity measurement of the entity.
Further, the reordering method in step 4 preferentially selects a label different from the existing label and a label complementary to the existing label, so as to generate a final label set.
Further, when reordering, calculating
Figure BDA0002678257290000024
Sorting is performed, wherein the candidate label liTo be in a candidate tag set Lc tBut not in the final set of labels LtIn d (l)i) Is aiIs scored for discrimination, reward (l)i，Lt) Is aiPotential contribution to coverage increase of a positive case entity in the knowledge-graph, where penalty is liTo LtThe potential impact of increased redundancy. Formula (II)
Figure BDA0002678257290000025
And
Figure BDA0002678257290000026
where reward and dependency are defined, δ is the bias factor, εtIs a set of entities of type t,
Figure BDA0002678257290000031
representative Label liThe sample entity of (1).
Compared with the prior art, the invention has the following advantages and beneficial effects:
the entity portrait method based on the knowledge graph focuses on finding the labels with the distinguishing performance and fully embodies the distinguishing performance among the entities; the entity tag distinguishing performance is deeply researched through an HAS model and a reordering method, the redundancy in a tag space is reduced, the entity similarity measurement is more comprehensive, and the tag distinguishing capacity is improved. The invention provides the research of the entity portrait in the knowledge map for the first time, which can be a good supplement to the research of the entity abstract and is beneficial to the realization of downstream tasks such as entity linkage, entity recommendation and the like.
Drawings
FIG. 1 is a flow chart of solid image rendering according to the present invention.
Fig. 2 is a diagram of three strategy for searching HAS in the present invention.
FIG. 3 is a diagram illustrating an example of a physical image according to the present invention.
Detailed Description
The technical solutions provided by the present invention will be described in detail below with reference to specific examples, and it should be understood that the following specific embodiments are only illustrative of the present invention and are not intended to limit the scope of the present invention.
For convenience of understanding, terms referred to in the embodiments of the present application are explained below:
knowledge Graph (Knowledge Graph): given a knowledge graph
Figure BDA0002678257290000032
Wherein
Figure BDA0002678257290000033
To represent
Figure BDA0002678257290000034
The set of nodes in (1), epsilon refers to the set of entities,
Figure BDA0002678257290000035
a set of texts;
Figure BDA0002678257290000036
representing a set of edges, each edge connecting one entity with another entity or text; τ represents a type function that acts on entities, mapping each entity to one or more predefined types; μ denotes a label function, mapping each edge to an attribute.
Label and Label Set (Label and Label Set): given a
Figure BDA0002678257290000037
A type t, a tag set LtA finite set of tags representing features describing type t. L istIs a triplet: l ═ i<t,lproperty,lvalue>。lpropertyRefers to an attribute, which may be a unique attribute or relationship owned by an entity, and lvalueRefers to the attribute value.
The core of the invention is to construct a set containing discriminative labels for each entity type. Specifically, the schematic flow chart of the entity portrayal method based on the knowledge graph provided in the embodiment of the present application is shown in fig. 1, and includes the following steps:
step 1: building a pool of candidate tags
The entity categories include: airlines, bands, baseball players, lakes, universities, philosophies, songs, political parties, television shows, comedians, academic journals, actors, books, mountains, radio stations, and movies, each entity category comprising a number of entities. The characteristics of the entities are heterogeneous in structure, and in order to help a user to fully understand the distinctiveness of the entities, the tags are divided into attribute tags and relationship tags according to different characteristic structures of the entities, wherein the attribute tags comprise attribute interval tags and attribute value tags, and the relationship tags comprise relationship-entity tags and relationship-attribute tags. Under the condition of no prior knowledge, all tags are violently enumerated from the knowledge graph in a mode of automatically generating the tags. By enumerating all combinations of attributes and attribute values or combinations of relationships and entities, candidate attribute value tags and relationship-entity tags can be directly generated, and generation of candidate attribute interval tags and relationship-attribute tags is complex. The continuous value of the attribute is generated to include a wider interval of the value, such as < Film, rating, [8.0,9.0] >.
Finding a suitable interval for the tag is a crucial step in generating the attribute interval tag, which is essentially a continuous numerical discretization problem of the tag. Simple discretization rules are set to find suitable intervals of specific values, for example, using a constant-width discretization method using a five-year period as the interval of various years. For other types of numerical values, a discretization algorithm based on local density is adopted, and the main idea is to find out a density interval of the attribute values, so as to ensure that the density in the middle of the interval is high and the density near the boundary is low. After the attribute values are ordered, the density values show a multi-peak phenomenon, and each peak value of the density distribution represents a boundary between two intervals.
Step 2: filtering candidate tags
By performing the preliminary screening on the tags in step 1, the candidate pool may contain many low-quality tags, which provide very limited or even misleading information. These low quality candidate tags are classified into two categories, one category is tags without representativeness, for example, in the Drug bank, there is a candidate tag < Drug, access id, "DB 0031600" >, but this tag can only describe one entity (a Drug called acetominophen) but cannot represent other entities. Another source of unrepresentative labels is noisy data. In DBpedia, the birthday of some soccer players was 2915, and these incorrect features would result in a nonsensical tag. The other is no distinguishing label. If most entities in a knowledge graph have a common characteristic label, the user cannot distinguish the entities by the label, which provides an almost zero amount of information in the understanding of the entities.
A simple heuristic rule is used to filter out these two types of low quality labels. Given an entity type t and candidate labels l associated with t, we willtIs defined as the set of all entities under type t, and
Figure BDA0002678257290000041
representing a set of regular entities associated with tag l, i.e.
Figure BDA0002678257290000042
support (l) represents the ratio of the positive case, and α and 1- α represent the upper and lower bounds, respectively. We assume that tags with low support values tend to be unrepresentative, while high support values represent no distinctiveness.Therefore, the tags in the middle are considered to be high quality candidate tags. The positive case is defined as an entity that matches the label, while the negative case is defined as an entity that does not match the label. And obtaining a candidate tag set after screening, and waiting for further discrimination evaluation.
And step 3: evaluating candidate tag distinctiveness
After generating candidates, all tags require further evaluation to ensure that the final tag results for the entity are discriminative, with a clear boundary between positive and negative examples. Use of
Figure BDA0002678257290000043
And
Figure BDA0002678257290000051
to measure distinctiveness. Wherein the entity i, j corresponds to the type t through the type mapping function tau,
Figure BDA0002678257290000052
for the sake of a positive example,
Figure BDA0002678257290000053
is a negative example; for a tag/of the type t,
Figure BDA0002678257290000054
show a positive example of
Figure BDA0002678257290000055
A negative example is shown. We define d (l) as the degree of distinction of l, i.e.
Figure BDA0002678257290000056
Average internal similarity of
Figure BDA0002678257290000057
And
Figure BDA0002678257290000058
the average external similarity of (a). sim (i, j) represents the similarity between entities i and j.
A multi-modal entity representation model is proposed to measure the similarity of entities, called HAS model. The embedded representation of each entity is learned by the HAS, keeping the entities closed in a continuous low-dimensional space if they share one or more structural patterns. HAS considers three structural models: the model simplifies the operation of entity representation, and is an efficient entity representation method for evaluating the distinguishability in a large-scale knowledge graph.
Taking the knowledge graph as input, and performing path searching operation on each entity to obtain a group of paths starting from the entity. Based on three strategies of HAS, three types of paths are obtained: (1) the H strategy is used for finding an H path representing a homogeneity pattern; (2) the strategy A is used for finding an A path representing the equivalence of the attributes; (3) the S policy is used to find an S path that represents a structural equivalence. Each type of path reflects some aspect of the structural pattern, and after path finding, the paths will be proportionally mixed as a feature of the entity. Finally, we learn the feature representation using a random walk model. Fig. 2 illustrates three path finding strategies. The top half of the graph represents the input KG portion, where nodes are entities, white rectangles are textual information, and edges are relationships or attributes connecting the entities and the text. Entities are labeled as nodes of different styles, depending on the type. The lower part of the diagram shows three path finding strategies from entity x. The explanation for each strategy is as follows:
and (4) strategy H: a simple random walk strategy to find H-paths reflecting homogeneity patterns, i.e. direct connections between entities. In fig. 2(a), starting from x, a plurality of paths are generated by Depth First Search (DFS).
Strategy A: for finding an a-path reflecting the equivalence of the attributes. Starting with x, strategy A attempts to find subsequent entities of the same type in the knowledge-graph that most closely resemble the x attribute values. In fig. 2, the entities x and y, z are all of the same type. It can be seen that z is more similar to x than y, since z is the same gender as x, and z is more closely aged to x than y. So, based on the policy of attribute equivalence, z is selected as the successor node to the path from x.
Fig. 2(b) shows a random walk model for finding a path, we first embed the same type of entity into an attribute space. For each type t, its attribute space has
Figure BDA0002678257290000059
The ratio of vitamin to vitamin is,
Figure BDA00026782572900000510
the number of attributes representing t. In the example, x, y, and z are embedded in one 2D space (age and gender). After normalization processing is carried out on each dimension, the hypercube taking x as the center in the multidimensional space depicts the neighborhood of x. We define the side length of the hypercube as 2r and entities falling within the region are considered neighbors of x. In the a path, one of the neighbors is randomly selected as the successor entity to x. The iteration continues until a fixed length a path is generated for x. The initial setting of 2r is estimated based on the average spacing between adjacent entities in attribute space. The hyper-parameters may enlarge or reduce the hypercube to bring the number of neighbors in the hypercube close to the average number of direct neighbors in the original knowledge-graph.
S strategy: structural equivalence is typically embedded in the local structure of a entity. For example, if two professors play similar roles in their social networks, they have a high similarity in structure, e.g., each of them has connections to many students. Similar to the a policy, the S policy finds the S path by embedding entities in the fabric space. Given a type t, its structure space has
Figure BDA0002678257290000061
Vitamin A, wherein
Figure BDA0002678257290000062
Is the number of types in the knowledge graph. The component of an entity in a certain dimension t 'is the number of direct neighbors of type t'. In the figure2(c), the horizontal axis represents the number of neighbors, the type of which is marked with dot hatching, and the vertical axis represents the number of neighbors, the type of which is marked with diagonal hatching. The coordinates of x, y, and z are (2, 6), (3, 5), and (3, 4), respectively. The following path finding step is similar to the a strategy.
Path mixing: finally, the HAS model generates 3 sets of random walk paths for each entity: pH、PA、PSRespectively, representing the H-path, a-path and S-path, which will be sampled into the final feature set. As in the formula P ═ λHPH∪λAPA∪λSPSAs shown, P is the final feature set, λH、λA、λSIs a scaling parameter for sampling the path. A uniform path sampling strategy is λH:λA:λS1:1: 1. Biased sampling is a strategy for unbalanced weighting schemes. In particular, when λH:λA:λSThe HAS model is equivalent to DeepWalk as 1:0: 0. After path mixing, we follow the skip-gram learning process of Deepwalk, resulting in an embedded representation of the entity. Finally, the cosine similarity is used for similarity measurement of the entity.
For a distinguishing label, the positive case is similar to the negative case, and the negative case is different. The discrimination of the labels is evaluated by calculating the similarity between positive examples and the similarity between positive and negative examples. The more similar the positive case is, the greater the difference between the positive case and the negative case is, the more discriminative the tag is. In this step, the difference between positive and negative examples is measured by the HAS model, and the distinguishing label is retained.
And 4, step 4: reordering
Finally evaluating the candidate tags by using a reordering method, wherein the requirement of (1) is that the redundancy brought to the tag set is small; (2) the integrity of the labelsets is improved. The first requirement is to preferentially process tags that are different from existing tags, and the second requirement is to favor tags that are complementary to existing tags.
Such as formula
Figure BDA0002678257290000063
As shown, given a set of candidate tags Lc tBut not in the final set of labels LtCandidate tag l in (1)i，d(li) Is aiIs scored for discrimination, reward (l)i，Lt) Is aiPotential contribution to coverage increase of a positive case entity in the knowledge-graph, where penalty is liTo LtThe potential impact of increased redundancy. Formula (II)
Figure BDA0002678257290000064
And
Figure BDA0002678257290000065
where reward and dependency are defined, δ is the bias factor, εtIs a set of entities of type t,
Figure BDA0002678257290000066
representative Label liThe sample entity of (1). Finally, based on liAnd sequencing the candidate labels by the calculated values, and adding the candidate labels into the label set one by one to generate a final label set. This step can effectively reduce redundancy in the tag space.
And 5: visual generation of entity portrait results
All entity descriptions will be traversed to find if an entity matches certain tags. Eventually, the visualization of the entity tag will be presented to the user to help quickly understand the uniqueness of the entity. Two physical portrait results of the present application are shown in FIG. 3. (a) The entities in the graph are movie entities, namely, lyon, defined in a movie knowledge graph LinkedMDB; (b) the entity in the figure is a band entity, beastieBoys, defined in DBpedia. Each entity has five tags, each tag extracted from KG and labeled in dark grey, where "≠ 80%" indicates that the entity is different in this characteristic from other 80% of movies or bands; "> 60%" or "< 95%" indicates that the entity has a larger or smaller value in the feature than other movies or bands.
The technical means disclosed in the invention scheme are not limited to the technical means disclosed in the above embodiments, but also include the technical scheme formed by any combination of the above technical features. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and such improvements and modifications are also considered to be within the scope of the present invention.

Claims (8)

1. A novel knowledge graph entity portrait method is characterized by comprising the following steps:
step 1: building a pool of candidate tags
Giving a knowledge graph as input, automatically enumerating all possible labels for each entity type, and putting the labels into a label pool;
step 2: filtering candidate tags
Filtering the low-quality labels through a filter to obtain a candidate label set;
and step 3: evaluating candidate tag distinctiveness
Measuring the distinguishability of the label through a uniqueness evaluator, calculating the similarity between positive examples of entities matched with the label and between positive examples and negative examples of entities not matched with the label, and evaluating the distinguishability of the label;
and 4, step 4: reordering
Selecting a label with distinctiveness by a reordering method to generate a final label set;
and 5: visual generation of entity portrait results
And traversing all entity label descriptions, searching whether a certain entity is matched with certain labels or not, obtaining a final portrait result, and visually presenting the portrait result to a user.
2. The novel knowledge graph entity representation method as claimed in claim 1, wherein in the step 1, the tags comprise attribute tags and relationship tags, the attribute tags comprise attribute interval tags and attribute value tags, and the relationship tags comprise relationship-entity tags and relationship-attribute tags; the process of generating the label is as follows: tags are automatically generated by violently enumerating all attributes and combinations of attribute values, or combinations of relationships and entities, from the knowledge-graph.
3. The method for representing an entity of knowledge graph as claimed in claim 2, wherein when the attribute interval label is generated, the continuous value of the attribute is discretized, and the appropriate interval is divided by a method combining the equal-width discretization and the density discretization.
4. The novel knowledge-graph entity representation method as claimed in claim 1, wherein in step 2, heuristic rules are used to filter low-quality labels, including labels without representativeness and labels without distinctiveness.
5. The novel knowledge-graph entity representation method as claimed in claim 4, wherein the heuristic rules specifically include:
for an entity type t and candidate labels l associated with t, ε istIs defined as the set of all entities under type t, and
Figure FDA0002678257280000011
Figure FDA0002678257280000012
represents a set of proper instances of the entity associated with tag l, the ratio support (l) of the proper instances when matching the tag l satisfies
Figure FDA0002678257280000013
Then, the label is considered as a high-quality candidate label, and alpha and 1-alpha respectively represent an upper bound and a lower bound; when support (l) is out of the upper and lower bounds, it is considered a low quality tag.
6. A novel knowledge-graph entity portrait method as claimed in claim 1, wherein said step 3 employs a multi-mode entity representation model to measure the similarity between entities, and takes into account the homogeneity mode, the attribute equivalence mode and the structure equivalence mode to generate 3 random walk paths; then, mixing the 3 paths according to a certain proportion, and following the skip-gram learning process of deep walk to obtain the embedded representation of the entity; finally, the cosine similarity is used for similarity measurement of the entity.
7. The novel method for representing an entity by knowledge graph as claimed in claim 1, wherein the reordering method in step 4 selects preferentially the tag different from the existing tag and the tag complementary to the existing tag to generate the final tag set.
8. The method of claim 7, wherein the computing is performed during reordering
Figure FDA0002678257280000021
Sorting is performed, wherein the candidate label li isIn the candidate tag set Lc tBut not in the final set of labels LtIn d (l)i) Is aiIs scored for discrimination, reward (l)i，Lt) Is aiPotential contribution to coverage increase of a positive case entity in the knowledge-graph, where penalty is liTo LtThe potential impact of increased redundancy. Formula (II)
Figure FDA0002678257280000022
And
Figure FDA0002678257280000023
where reward and dependency are defined, δ is the bias factor, εtIs a set of entities of type t,
Figure FDA0002678257280000024
representative Label liThe sample entity of (1).
CN202010954910.9A 2020-09-11 2020-09-11 Novel knowledge graph entity portrait method Pending CN112364171A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010954910.9A CN112364171A (en) 2020-09-11 2020-09-11 Novel knowledge graph entity portrait method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010954910.9A CN112364171A (en) 2020-09-11 2020-09-11 Novel knowledge graph entity portrait method

Publications (1)

Publication Number Publication Date
CN112364171A true CN112364171A (en) 2021-02-12

Family

ID=74516802

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010954910.9A Pending CN112364171A (en) 2020-09-11 2020-09-11 Novel knowledge graph entity portrait method

Country Status (1)

Country Link
CN (1) CN112364171A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113885862A (en) * 2021-09-29 2022-01-04 武汉斗鱼鱼乐网络科技有限公司 Head photo frame multiplexing method, storage medium and electronic equipment

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103577549A (en) * 2013-10-16 2014-02-12 复旦大学 Crowd portrayal system and method based on microblog label

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103577549A (en) * 2013-10-16 2014-02-12 复旦大学 Crowd portrayal system and method based on microblog label

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
张祥等: "基于知识图谱的实体标签可视化", 指挥信息系统与技术, vol. 11, no. 3, 28 June 2020 (2020-06-28), pages 1 - 9 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113885862A (en) * 2021-09-29 2022-01-04 武汉斗鱼鱼乐网络科技有限公司 Head photo frame multiplexing method, storage medium and electronic equipment

Similar Documents

Publication Publication Date Title
US9256670B2 (en) Visualizing conflicts in online messages
Steiniger et al. An approach for the classification of urban building structures based on discriminant analysis techniques
EP2192500B1 (en) System and method for providing robust topic identification in social indexes
Denizci Guillet An evolutionary analysis of revenue management research in hospitality and tourism: is there a paradigm shift?
CN107835113A (en) Abnormal user detection method in a kind of social networks based on network mapping
Koylu et al. Design and evaluation of line symbolizations for origin–destination flow maps
De Nisco et al. From international travelling consumer to place ambassador: Connecting place image to tourism satisfaction and post-visit intentions
Centobelli et al. Mapping knowledge management research: A bibliometric overview
CN111222847B (en) Open source community developer recommendation method based on deep learning and unsupervised clustering
Wilkens Digital humanities and its application in the study of literature and culture
US20220092108A1 (en) Determining fine-grain visual style similarities for digital images by extracting style embeddings disentangled from image content
Silver et al. A Markov model of urban evolution: Neighbourhood change as a complex process
Chen et al. An extended study of the K-means algorithm for data clustering and its applications
Nguyen et al. A bibliometric analysis of research on tourism content marketing: Background knowledge and thematic evolution
Cho et al. Classifying tourists’ photos and exploring tourism destination image using a deep learning model
Sigler et al. Socio-spatial relations observed in the global city network of firms
Yuan et al. Point cloud clustering and outlier detection based on spatial neighbor connected region labeling
CN112364171A (en) Novel knowledge graph entity portrait method
Burns et al. Towards qualitative geovisual analytics: a case study involving places, people, and mediated experience
Qiu et al. Past, present, and future of tourism and climate change research: bibliometric analysis based on VOSviewer and SciMAT
Chen et al. Using Network Embedding to Obtain a Richer and More Stable Network Layout for a Large Scale Bibliometric Network
Pang et al. Salient object detection via effective background prior and novel graph
Mauleón-Méndez et al. Tourism research: A bibliometric and country analysis
Courtois et al. Computer Vision and Internet Meme Genealogy: An Evaluation of Image Feature Matching as a Technique for Pattern Detection
Bizzoni et al. Predicting Literary Quality How Perspectivist Should We Be?

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination