table of Contents
1.1.3 relationship with the body of knowledge map
1.2 Knowledge of the significance map
1.2.1 Knowledge Mapping is the cornerstone of cognitive intelligence
1.2.2 knowledge to guide one become an important way to solve the problem
1.3 Value of Knowledge Mapping
1.3.3 Intelligent Recommendation
1.3.4 natural human-computer interaction
1.4 Future research directions
1.1 Basic concepts:
Knowledge Mapping is a narrow class of knowledge representation, is essentially a large-scale semantic network. Mapping knowledge in a broad sense is a general term for a series of big data era of knowledge-engineering technology, big data refers to the emerging discipline of knowledge engineering to some extent.
1.1.1 narrow concept
1. Knowledge Mapping as a connotation Semantic Web
As a knowledge representation, knowledge map is a large-scale semantic network, including entity (entity), the concept (concept) and various semantic relationships between. Two points: First, it is a semantic network, which is the essence of knowledge map; the other, in fact, large-scale, which is the fundamental difference between traditional knowledge map and Semantic Web.
|
FIG composition of semantic networks (asterisk indicates there may be a plurality of different attributes or relations) |
Semantic network is a way to graphically (Graphic) in the form of knowledge representation by vertices and edges, which are constituent elements substantially and edges. Shown below is a typical semantic network. Semantic network entity may be a point, the concepts and values:
- Entity . Entity or instance of an object is sometimes referred (instance). Entity is the basis for the existence of the property, and must be free, that is independent and not dependent on other things exist.
- Concept . The concept is also known as category, class (category or class) and so on. Such as "philosopher" does not refer to a particular philosopher, but rather refers to a class of people, this group of people have the same description templates, constitute a class or concept. The concept corresponding to the verb "conceptualized (conceptualize)" or "Categorization (Categorize)." Conceptualization generally refers to the process in the recognized text related concepts.
- Value . Each entity has a certain property value. Property value may be a common value type, date type or a text type.
Edge knowledge map can be divided into property (property) and the relationship (relation) categories. Attribute describes the effects of certain aspects of the entity, such as a person's date of birth, height, weight and so on. Attribute is that people recognize the world, describing the foundation of the world. Relationship can be considered as a special kind of property, the property value when an entity is an entity when, in essence, is the property relations.
2. Knowledge of the advantages and disadvantages map
Semantic Web and traditional knowledge map What is the difference? The answer to this question determines the existence of the value of knowledge map. On a huge scale mapping knowledge, in addition, it is also reflected in its semantically rich, excellent quality, structure and friendship (triplet) and other characteristics: knowledge and traditional semantic network map the most obvious difference is reflected in the scale.
Knowledge Mapping change in size also determines the knowledge map significant differences exist in knowledge acquisition to knowledge and applications from traditional semantic network. These differences constitute a unique challenge to build knowledge and application profiles, respectively, are discussed below:
- Lack of high-quality mode . Enhance the knowledge map scale will often pay for quality. Building knowledge map original intention was to adapt to the knowledge needs in an open environment. In order to allow more storage of knowledge is bound to properly relax the requirements for the quality of knowledge. Traditional database and knowledge base for data or knowledge which has a strict definition, data warehousing can have strict constraints. Almost all are susceptible to strictly defined exceptions. Therefore, mapping knowledge in the design mode usually take a kind of "economic, pragmatic" approach: that is, to allow mode (schema) definition of imperfect, or even missing. Schema definition incomplete or missing data semantic understanding and knowledge map data quality control challenges.
- Closed world assumption is no longer valid . The application of traditional database and knowledge base are usually established in a closed world assumption (closed world assumption, CWA) basis. CWA assumes that the database or knowledge base (or observed) the fact that the established facts do not exist. Obviously, this is a strong assumption applies only to closed areas. Most open the application does not comply with this hypothesis. In other words, does the fact that in these applications or knowledge is not necessarily false. For example, it is difficult to ensure that the Knowledge Graph information about Plato's complete, it may be missing information Plato parents. But common sense tells us that Plato must have parents. Non-compliance with the CWA to the application of knowledge map tremendous challenge.
- Large-scale automated knowledge acquisition become a prerequisite . Knowledge map is huge, its implementation relies on large-scale automated knowledge acquisition. Traditional knowledge engineering experts rely on knowledge acquisition is completed, this approach is difficult to achieve large-scale acquisition of knowledge, can not meet the size requirements of the knowledge map. Large-scale automated knowledge acquisition and knowledge is one of the fundamental difference with the traditional pattern of semantic networks. It should be noted that the large-scale automated knowledge acquisition approach is diverse, can be automatically extracted from the text, can be labeled based on knowledge of large-scale crowdsourcing platform can also be mixed in various ways. But no matter what kind of specific implementation, large-scale knowledge acquisition knowledge map construction are necessary.
1.1.2 broad concept
Knowledge Mapping technology development today, its meaning has been far beyond the scope of the Semantic Web, in practice it has been given more and more rich connotations. Today, under the more realistic scenarios, mapping knowledge as a technical system, the sum of a series of representative on behalf of big data technology knowledge engineering.
1.1.3 relationship with the body of knowledge map
Body (Ontology) described explicitly shared conceptual model [1], describe the relationship between concepts and concepts; Semantic Web is the key technology for adding semantics Semantic Web pages and Web ontology concept of mapping knowledge, two. concerned - describes the relationship between ontology and concepts, knowledge map mode is most layer is a conceptual model and logical basics of map atlas with the body of knowledge in common is that: both are defined by metadata support semantic service except that: knowledge map more flexible, dividing things by adding a label to support custom categories body focused description of the conceptual model, can represent knowledge of general, abstract description, the emphasis is conceptual. and the relationship between the concepts. most of the body that does not contain too many instances, filling ontology instances are usually carried out after the completion of construction of the body. knowledge Mapping is more focused describe the entity relationship, entity level of the body in a large number of rich and expansion . It is believed that the body is the expression of abstract knowledge map, describing the upper pattern mapping of knowledge; knowledge is a body map Instantiation is based ontology knowledge.
[1]Gruber TR. A translation approach to portable ontologyspecifications. Knowledge Acquisition, 1993, 5(2): 199–220.
( Huang Hengqi , to Juan, Liao, etc. Summary of mapping knowledge [J]. Computer System, 2019,28 (. 6): 1-12 the DOI:. 10.15888 / j.cnki.csa.006915. )
(Reference books) ontological philosophy stems from the body, focused on the presence of provisions and characterization. An important motivation for body area is artificial intelligence, knowledge sharing and reuse, and interoperability of data. Different autonomous systems (such as different sites, different machines) only to follow the same "world view", it may form a similar "understanding." The Semantic Web ontology developed a lot of standard definition language and resource exchange. Thus, the computer field is focused on the body frame conceptual expression of cognition, semantic relationship between the expression of concept, often accompanied axiomatic characterization system concepts.
Body portrait of people perceive a frame extension of the present art. Like the relationship between human flesh and blood and bone marrow relationship between the frame and the examples. No frame, unable to support the understanding of the world Du machine or a particular area, the frame is the heart and soul of cognition. But not only the framework instance, than like a good spirit but weakness can not achieve machine intelligence. As defined in the machine body, like our view of the world passed to the machine. Obviously, this work requires a human expert to complete. In comparison, knowledge-rich maps of the entity and relationship instances. In the early construction of the knowledge map, the schema definition is essentially complete body in the definition of the task.
1.2 Knowledge of the significance map
1.2.1 Knowledge Mapping is the cornerstone of cognitive intelligence
The so-called cognitive intelligence is the ability to let the machine the human perception of the world. Intelligent machines cognitive ability is the core of this "understanding" and "explanation", both of which are closely related with the knowledge map. First, we need to machine "understanding" and "explanation" put forward an explanation. Machine understand the nature of the data from the data is mapped to the elements of knowledge Knowledge Graph (including entities, concepts and relationships) of. Not difficult to find through the process of reflection of human understanding of the text, "understanding" can be regarded as established from the data (including text, images, voice, video, data, etc.) to process mapping between the Knowledge Graph entity, the concept of property. The "explanation" is the talk of knowledge and issues in the Knowledge Graph or associated data.
Cognitive mapping knowledge of the importance of intelligent machines is also reflected in the following areas:
- Knowledge Mapping enabled machine language cognition.
- Knowledge Mapping enabling interpreted artificial intelligence.
- Knowledge helps to enhance the ability of machine learning.
1.2.2 knowledge to guide one become an important way to solve the problem
Knowledge map for achieving an important role in cognitive machine intelligence, knowledge determines the guide will become one of the main ways to solve the problem. Moment, take the computer to solve the problem mainly data-driven approach, which is to establish a statistical model from sample data, statistical mining law to solve the problem. To enhance the effect, data-driven approach typically requires more sample data. However, even if the data sample size is bigger, purely data-driven approach still faces the effects of the "ceiling."
And to break the "ceiling", requires knowledge guide. Many knowledge-intensive applications for knowledge guide made a strong appeal. For example, in the penalty prediction problem the company demands, the same two wounding statement of the case, is a suspect in advance with a dagger, the other is the suspect pick up a brick, even if all other statements exactly the same, it is entirely the result of the penalty different (the former will be determined as premeditated murder, which will be judged spur of the moment, sentencing results completely different). The reason, the penalty is fundamentally determined by the judicial knowledge. Data-driven approach purely statistical word frequency characteristics of the text, difficult to effectively solve practical tasks such knowledge-intensive. Practical applications require more and more data-driven and knowledge to guide the combination of a bottleneck effect purely data-driven approach to break through based on statistical learning.
Therefore, knowledge will become more important than data assets. If the data is oil, then knowledge is like extract than oil. If we get complacent value directly from the data, like direct output of oil profits. However, a more significant value inherent in its deep processing of oil extract. Oil extraction process and knowledge processing process is very similar, all have complex processes, are large-scale systems engineering.
1.3 Value of Knowledge Mapping
Is a process of liberation of the human brain continues to develop the process is essentially the only machine cognition. In the industrial revolution and the information age, physical strength is gradual emancipation of mankind; and with the development of artificial intelligence, especially the development of cognitive intelligence technology, the human brain will also be gradual emancipation. More and more knowledge work will gradually be replaced by a machine, accompanied by further enhance the productivity of the machine. Based on a broad and diverse spectrum of cognitive intelligence knowledge application.
1.3.1 Data Analysis
Precise and accurate analysis of large data requires knowledge map. Today, more and more industries or enterprises have accumulated large scale objective data, but these data do not achieve the desired value, many large data also consumes a lot of operation and maintenance costs. Big Data has not only create value, in many cases, become a sum of negative equity. The root cause of this phenomenon is that the current machines such as background knowledge map, can not accurately understand the data, limiting the accurate and precise analysis of large data, restricted cash value of big data. In fact, public opinion analysis, Internet business insights, as well as military intelligence analysis and business intelligence analysis, we need to do accurate analysis of large data, and this precise analysis must have a strong background knowledge to support.
In addition to accurate big data analysis, another important trend in the field of data analysis - fine analysis, but also intelligent and cognitive knowledge map made demands. For example, many car manufacturers want to personalize manufacturing, that want to collect user feedback and evaluation of the car from the Internet, and as a basis to achieve auto-demand and customization.
1.3.2 wisdom Search
First of all, accurate search intent to understand. For example, Taobao search for "Huawei charger," the user's intention is obviously to search for a charger, rather than a Huawei phone, this time Taobao should feedback to the user a number of products to choose from the charger.
Second, the search target complex and diversified. Traditional text-based search for objects, more and more applications in the future want to search for pictures and sound, and even search code, video, design material, etc., anything can be searched.
Again, the search size diversity. Now not only do the search chapter-level search, but also hope to do a paragraph level, sentence level, vocabulary level of search. Most of the traditional knowledge management can do document-level search, knowledge management such coarse-grained been difficult to meet the practical application of fine-grained knowledge acquisition needs.
Finally, cross-media cooperative search. In traditional search for elemental mostly single data source search, information retrieval difficult to meet the needs of users. For example, search for text difficult With video, picture, image search mainly for the use of its own picture information, for utilization of a large number of text messages is not high. Coordinate cross-media search demand is increasing.
Knowledge-based intelligent search map is based on the long tail of search, the search engine in the form of knowledge of the search result card to show up. After the user's query request inquiry Semantic understanding and knowledge retrieval in two stages : 1) type query semantic understanding. Mapping knowledge semantic analysis of the query type include: ① to requests for text segmentation, POS tagging, and error correction; ② described normalization, so as to match the [114] and relevant knowledge in the knowledge base; ③ Analysis Context . In a different context, user query where the objects will be different, and therefore requires a combination of knowledge at the time of the emotional spectrum user, the user will need to answer at this time timely feedback to the user; ④ query expansion. Query intent and clear concepts of users, you need to join concepts in the context of the current expansion. 2) Knowledge Retrieval. After the standard query into the query expression analyzing repository search engine, the engine will retrieve entities in the Knowledge Base and its category, high aspect relationship and correlation matching entity [115]. Deep mining and refining through the knowledge base, the engine will be given with full knowledge of the importance of the sort. Intelligent Search Engine demonstrate knowledge in three main forms: 1) semantic data integration. For example, when someone searches for Van Gogh, the search engine will give a detailed knowledge of the card Van Gogh's life, and in line with pictures and other information; 2) gives the user direct query answer to the question. For example, when a user searches for "How tall is Yao Ming?", The results of the search engine is "226 cm"; 3) gives the recommended list based on a user's query [7] and so on.
(Xuzeng Lin, Sheng Yong Pan, He Lirong, etc. Survey of Knowledge Mapping. University of Electronic Science and Technology, 2016, 45 (4): 589 -606 [doi: 10.3969 / j.issn.1001-0548.2016.04.012]. )
1.3.3 Intelligent Recommendation
First, the scene of the recommendation. Any keyword search, any opinion behind the shopping cart of goods, embodies specific consumer intentions, is likely to correspond to a specific consumer scene. Create the scene map for precise recommendations based on the scene map, it is critical for the electricity supplier is recommended.
Second, under the recommended cold start phase. The use of external knowledge from the knowledge map, especially with regard to knowledge of the user and items to enhance the user with an item description, improve matching accuracy, the system is an important idea to spend the cold start phase as soon as possible.
Third, cross-cutting recommendation. There is a large number of heterogeneous platforms on the Internet, to achieve cross-cutting between the platform has recommended more and more applications.
Fourth, knowledge-based content recommendation. If a user searches for milk powder in electronic business platform, so users recommend some parental knowledge about the prevention of infant formula and the amount of water needed daily, a common disease. Knowledge of these recommendations will significantly enhance the user receiving a degree of trust and recommended content. Content and knowledge needs behind consumer behavior will be an important consideration recommended.
1.3.4 natural human-computer interaction
Another very important form of intelligent systems is a natural human-computer interaction. Human-computer interaction will become more and more natural, more simple. The more natural, simple way to interact with the machine requires a strong intelligence. Natural human-computer interaction, including natural language questions and answers, dialogue, somatosensory interactive, face interactions. Natural language interaction requires that the machine can understand natural human language. Conversational interaction (conversation UI), interactive question and answer (QA) will gradually replace the traditional keyword search-based interaction. Another important trend is that anything can quiz.
Q system (Question Answering, QA) refers to allow computer users to automatically answer the questions raised, is an advanced form of information service. Unlike existing search engine, the system returns the user's Q & A is no longer relevant documents sorted based on keyword matching, but accurate natural language answer. Professor Etzioni, director of the Turing Center at Washington University in 2011 had published an article "Search Needs a Shake-Up" on Nature, which clearly states: "direct and accurate way to answer user questions in natural language question answering system will constitute the next generation of search the basic form of the engine "[Etzioni O., 2011]. Therefore, the question answering system was seen as one of the next disruptive technology information services, is considered to be one of the main means of verification machine has the ability to understand the language.
1.3.5 Decision Support
Knowledge maps provide deep relationship discovery and reasoning ability for decision support. People are increasingly satisfied with the "so and so is the wife of so and so" simple association of such a discovery, and hope the discovery and excavation of some deep, hidden relationships. In the financial sector, we are very concerned about the possible investment relations, for example, why an investor to invest in a company; we are very concerned about the financial security, such as credit risk assessment requires analysis of a lender of connected characters and the associated company's credit rating. Therefore, a knowledge map contains a variety of semantic association, digging deeper relationships between entities, it has become an important adjunct to decision analysis.
Decision support in the military field is reflected in?
By evaluating the situation on the battlefield, the battlefield to establish entity-relationship database, the use of big data analytics technology to provide scientific decision-making aid.
1.4 Future research directions
Knowledge map provides a new way to achieve representation, storage and management of knowledge, growing interest and research to get some progress. This paper introduces the basic aspects of the knowledge map of the building, storage, and applications on the elaborated knowledge the relationship between the map and the analysis of the existing body of knowledge map at home and abroad, showing that research knowledge map has a certain achievements, future research is:
Question (1) distributed memory mapping of knowledge. Because of the special nature of the structure of the knowledge map, with the increasing amount of data, how it is distributed storage a practical significance. The problem includes how to conduct knowledge map reasonable divided storage without affecting its function, load balancing knowledge map, knowledge map memory model.
(2) Knowledge Mapping reasoning. Knowledge Mapping build reasoning to support knowledge map constructed by detecting the presence or absence of prior knowledge and the discovery of logical contradiction unknown from the known relationship between knowledge, to ensure the consistency and integrity of the knowledge map , but also to enrich and expand the knowledge map. On the other hand, knowledge map application reasoning is also of concern. by adding rules in the field of knowledge, knowledge map application reasoning inferences can be achieved in the field of knowledge, automated decision aid, intelligent questions and answers and forecasts Wait.
(3) At present, most of the knowledge spectrum utilization and reuse rate is not high, even it was shelved after the completion of building work; on the other hand, there are the actual needs of enterprises is the lack of channels for building knowledge map for this. case, the future can be considered:. ① strengthen the knowledge map and ontology building system engineering theory of knowledge, and training of relevant personnel ② increase the general knowledge of map building efforts, and industry knowledge map only when the actual demand and then according to the situation building. ③ continue to strengthen research and development of knowledge maps and automatic ontology construction method , improve the automation of the build process .
references
"Knowledge Mapping: Concepts and Techniques" Xiao Yang Hua, eds;
Huang Hengqi , to Juan, Liao, etc. Summary of mapping knowledge [J]. Computer System, 2019,28 (6): 1-12 DOI :. 10.15888 / j.cnki.csa.006915.
