Knowledge Map Construction Process _ Knowledge Map

Briefly introduce the process of building a knowledge map: 1. Data Source: (Data level) Encyclopedia class data (Wikipedia semi-structured, freebase structured), structured data (common semantic datasets such as DBpedia and Yago, including MusicBrainz and DrugBank Knowledge of specific fields, semi-structured data, automated AVP (attribute-value pair) extraction, and search log mining, discovering the latest emerging entities, the process of learning bootstrapping methods based on bootstrapping collaborative patterns: Given a hand The seed NEs of a category c:learning context features of seeds from queries extracting new seed entities of Categor Y C using the learnt context features expanding context features using the expanded seed set #属性-value pair (Attribute-value pair , also known as AVP), is used to characterize the intrinsic properties of an entity, whereas a relationship (relation) is used to connect two entities to depict their associations.
2. From the Extraction Atlas (extraction Graphs) to the Knowledge Map: (1) Entity alignment (Object alignment), the clustering algorithm for multiple source data, the key is to define the appropriate similarity measure (2) Knowledge Map schema Construction, It is equivalent to establish ontology (Ontology), the most basic ontology includes concept, concept level, attribute, attribute value type, relation, Relation definition domain (field) concept set and relation Range (ranges) concept set. The Top-down approach refers to the Ontology of ontology through the Ontology editor (Editor), which is not a process from scratch, but relies on the pattern information extracted from the high-quality knowledge obtained from the encyclopedia class and structured data. The bottom-up approach incorporates the categories, attributes, and relationships found through the various extraction techniques described above, especially through search logs and Web table, and merges these high confidence patterns into the knowledge map. The merge process uses an alignment algorithm that resembles an entity alignment. (3) The resolution of inconsistency. Take precedence over facts extracted from highly reliable data sources such as encyclopedia or structured data.
3. Knowledge Map Mining: (1) reasoning, for attributes, for relational (2) Entity importance ordering, when a query involves multiple entities, search engines will select more relevant and more important entities to show. The correlation metric of the entity needs to be calculated online while querying, while the entity importance and query-independent can be computed off-line, search engine companies apply the PageRank algorithm to the knowledge map to compute the importance of the entity (3) related entity mining. Use a theme model (such as LDA) to discover the distribution of topics in a virtual document set. Each of these topics contains 1 or more entities, each of which is related to entities in the same subject. When a user enters a query, the search engine analyzes the subject distribution of the query and selects the most relevant topics.
4. Update and maintain the knowledge map. (1) The relationship between type and collection search engine companies also use automated algorithms to extract new types of information from a variety of data sources, and if a certain type of collection can be retained for a long period of time, a professional person will make and name the decision and eventually become a new type. (2) Structured site wrapper maintenance search engine will regularly check whether the site is updated, using the latest site wrapper for AVP extraction (3) Knowledge map Update frequency type corresponding to the case is often dynamic (4) public packet (crowdsourcing) feedback mechanism The user can correct the facts related to the entities listed in the knowledge card presented in the search results. When many users point to an error, the search engine will adopt and fix
5. The application of the knowledge map in the search (1) Query understanding search engines are not all attributes of an entity, but are automatically selected according to the query that is currently entered, by selecting the most relevant attribute and attribute values. When the entity to be displayed is selected, use related entity mining to recommend entities that other users may be interested in for further browsing (2) questions answer the knowledge map another innovation for search is to return the answer directly, not just the sorted list of documents. The search engine not only understands the entities and their attributes involved in the query, but also needs to understand the semantic information corresponding to the query. Search engine through efficient graph search, in the knowledge map to find links to these entities and attributes of the child map and convert to the corresponding graph query (such as SPARQL) SPARQL: is a query language for RDF http://www.w3.org/TR/rdf-sparql-query/