CN107273418A - A kind of across Noumenon property chain inference method based on cloud platform - Google Patents

Abstract

The present invention discloses a kind of across Noumenon property chain inference method based on cloud platform, including：(1) the knowledge mapping H of unified form is obtained using semantic interlink method；(2) attribute chain inference network C is obtained using knowledge mapping；(3) object and attribute are replaced with into corresponding id, forms knowledge mapping H ', attribute chain inference network C '；(4) MapReduce frameworks, the parallel inference according to C ' to knowledge mapping H ' carry out attribute chains are used, and changes renewal C '；(5) the reasoning results are preserved into hdfs, and added it in knowledge mapping H '；(6) circulation step (4) and (5), (7) merge the reasoning results of successive ignition generation on hdfs untill the reasoning new until not producing is true.This method can support the parallel inference across the mass knowledge of body, with very strong autgmentability, have good practical value for the application of extensive semantic data reasoning.

Description

A cross-ontology attribute chain reasoning method based on cloud platform

technical field

The invention relates to computer semantic reasoning technology, in particular to a cloud platform-based cross-ontology attribute chain reasoning method.

Background technique

With the continuous development of the semantic web, the semantic web description language OWL based on the resource description framework has been widely used in ontology modeling and reasoning in various fields, including life sciences, media information, semantic spatiotemporal data, and social networks. And so on, the semantic data in various fields also showed explosive growth. Take the Linked Open Data project as an example. It proposes the concept of Linked data. Its purpose is to call on people to publish existing data as semantically linked data, so as to interconnect different data sources. So far it has contained more than 295 data sources and 31 billion triple records.

There are many hidden complex relationships among these massive semantic data, and the potential semantic information can be obtained by inferring the existing semantic information. These hidden semantic relationships are of great significance in practice. For example, biomedical workers can use semantic reasoning methods to obtain drug associations to assist the development of new drugs, and website data analysts can use user information to reason and connect.

However, the existing semantic inference engines often lack good scalability and can only process small-scale ontology. With the continuous growth of OWL ontology data, the reasoning engine running in the above-mentioned stand-alone environment needs to load a large amount of ontology data. When performing OWL reasoning on large-scale cross-ontology data, there are problems such as memory overflow, insufficient computing performance and scalability, and traditional semantic reasoning machines have been difficult to handle such massive semantic information. On the other hand, some parallel reasoning techniques that have been proposed cannot effectively solve the problem of large-scale complex semantic reasoning. The field of semantic research urgently needs a high-performance reasoning engine that can handle complex semantic associations to change this dilemma.

Contents of the invention

In view of this, the present invention provides a cross-ontology attribute chain reasoning method based on a cloud platform. Compared with other methods, the present invention realizes efficient discovery of complex semantic association information existing in massive semantic data through attribute chain reasoning, and has strong expansion capability.

A method for cross-ontology attribute chain reasoning based on a cloud platform, comprising the following steps:

(1) Use the semantic linking method to fuse ontology data in multiple fields to obtain a knowledge graph H in a unified format;

(2) Use the knowledge graph to obtain a series of inference rules A expressing the relationship between entities in a single ontology, and use the Owl2 attribute chain primitive of the semantic ontology description language to model the complex relationship between cross-ontology entities and entities, Obtain multiple attribute chain inference rules B expressing the relationship between entities across ontology entities, and A and B form an attribute chain inference network C;

(3) Assign a corresponding attribute id number to each attribute object in the knowledge graph and attribute chain inference network C, and assign a corresponding entity id number to each entity object to form knowledge graph H' and attribute chain inference network C';

(4) Using the MapReduce framework, according to the attribute chain reasoning network C', perform attribute chain parallel reasoning on the knowledge map H', and modify and update the attribute chain reasoning network C';

(5) Save a series of results obtained by reasoning in step (4) into hdfs, and add them to the knowledge map H';

(6) Judging whether the reasoning result of this time is consistent with the reasoning result of the last time, if so, execute step (7); if not, jump to execute step (4);

(7) End the reasoning, merge the reasoning results generated by multiple iterations on hdfs, and remove the repeated triples in the reasoning results, and then restore the corresponding text triples according to the attribute mapping table and entity mapping table, and convert the result Returned as the final inference result.

The relationship between entities and entities in a single ontology can be expressed by attributes, and in cross-ontology reasoning, this simple relationship may evolve into a very complex chain relationship, and these reasoning relationships can be effectively described by using inference rules.

There is a semantic gap between heterogeneous and cross-domain ontologies. The semantic link method can be used to associate entities and relationships representing the same object in different ontologies for the next step of reasoning and implementation.

When the semantic linking method is used for semantic fusion of cross-domain ontology, the distance between entities is calculated by designing a variety of similarity feature functions, so as to carry out entity linking and fusion. The similarity feature function is:

Similarity(X,Y)=Jac(X,Y)+Cos(X,Y),

X, Y are the description description information of the two entities respectively, Jac(X, Y) represents the Jaccard similarity, Cossine(X, Y) represents the cosine similarity, when Similarity(X, Y) is greater than 0.8, the entity Link.

The attribute chain reasoning rule B can effectively express the reasoning relationship and provide rule input for the subsequent reasoning method. The attribute chain reasoning network C can effectively describe the possible relationship between entities across ontology. The complex reasoning process is simplified through the attribute chain and the attribute chain network, which can not only effectively simplify the reasoning complexity, but also improve the parallel effect of the reasoning process at the same time, and provide a basis for implementing distributed reasoning algorithms.

Use the id number to replace the corresponding text object and attribute object, which can greatly improve the efficiency of reasoning.

The concrete process of described step (3) is:

Build an attribute mapping table and assign an attribute id number to each attribute object;

Build an entity mapping table and assign an entity id number to each entity object;

Replace the attribute chain object in the attribute chain reasoning network C with the attribute id to form the attribute chain reasoning network C';

Traverse each triple in the knowledge graph H, and replace the corresponding head node, tail node and relationship node with the entity id and attribute id to form the knowledge graph H'.

In step (4), using the MapReduce framework for reasoning can greatly improve the possibility and scalability of reasoning. The specific implementation process is as follows:

(4-1) Map stage: take (line number, triplet) key-value pair as input, and output (link attribute id key value, triplet) key-value pair;

(4-2) Reduce stage: use the (link attribute id key value, triplet) key-value pair output in the Map stage as the input of this stage, fuse triplets with the same id key value, and output (_, new triplet groups or pending triples);

(4-3) Merge and update the adjacent attribute chain objects in the attribute chain reasoning network C′, and reassign a new id for the attribute chain objects;

(4-4) Check whether there is a new triple output, if so, jump to step (4-1), if not, output the reasoning result.

The traditional semantic reasoning method is based on a stand-alone machine, and has obvious defects in the face of massive semantic data across ontology; however, the cloud platform-based cross-ontology attribute chain reasoning method of the present invention takes advantage of the scalability of the cloud platform and can handle large-scale The specific advantages of semantic data are as follows:

(1) The present invention uses attribute chains and attribute chain networks to simplify the modeling of the complex cross-ontology reasoning process, which overcomes the semantic reasoning that traditional reasoners can only complete established reasoning rules, and greatly enhances the flexibility and usability of reasoning.

(2) MapReduce, as a large-scale data computing framework, facilitates the processing of parallel algorithms, and HDFS can also provide a storage basis for large-scale knowledge graphs. The present invention utilizes HDFS to store large-scale semantic data, and simultaneously uses the MapReduce parallel framework Perform parallel reasoning across ontology attribute chains, greatly improving processing speed.

Description of drawings

Fig. 1 is the flow chart of the cross-ontology attribute chain reasoning method based on cloud platform of the present invention;

FIG. 2 is a cross-ontology massive biological knowledge graph in Example 1.

detailed description

In order to describe the present invention more specifically, the technical solutions of the present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

Referring to Fig. 1, the cross-ontology attribute chain reasoning method based on the cloud platform of the present invention includes:

S01, using the semantic linking method to fuse ontology data in multiple fields to obtain a knowledge graph H in a unified format.

S02, using the knowledge graph to obtain a series of inference rules A expressing the relationship between entities and entities in a single ontology, and using the Owl2 attribute chain primitive of the semantic ontology description language to model the complex relationship between entities and entities across ontology, and get Multiple attribute chain reasoning rules B expressing the relationship between ontology entities and entities, A and B form attribute chain reasoning network C.

S03, assign a corresponding attribute id number to each attribute object in the knowledge graph and attribute chain inference network C, and assign a corresponding entity id number to each entity object to form knowledge graph H' and attribute chain inference network C '.

This step is specifically:

Build an attribute mapping table and assign an attribute id number to each attribute object;

Build an entity mapping table and assign an entity id number to each entity object;

Replace the attribute chain object in the attribute chain reasoning network C with the attribute id to form the attribute chain reasoning network C';

Traverse each triple in the knowledge graph H, and replace the corresponding head node, tail node and relationship node with the entity id and attribute id to form the knowledge graph H'.

S04. Use the MapReduce framework to perform parallel reasoning on attribute chains on the knowledge map H' according to the attribute chain reasoning network C', and modify and update the attribute chain reasoning network C'.

This step is specifically:

(4-1) Map stage: take (line number, triplet) key-value pair as input, and output (link attribute id key value, triplet) key-value pair;

(4-2) Reduce stage: use the (link attribute id key value, triplet) key-value pair output in the Map stage as the input of this stage, fuse triplets with the same id key value, and output (_, new triplet groups or pending triples);

(4-3) Merge and update the adjacent attribute chain objects in the attribute chain reasoning network C′, and reassign a new id for the attribute chain objects;

(4-4) Check whether there is a new triple output, if so, jump to step (4-1), if not, output the reasoning result.

S05, save a series of results obtained by reasoning in S04 into hdfs, and add them to the knowledge map H'.

S06, judging whether the reasoning result of this time is consistent with the reasoning result of the last time, if yes, go to S07; if not, go to S04.

S07, end the reasoning, merge the reasoning results generated by multiple iterations on hdfs, and remove the repeated triples in the reasoning results, and then restore the corresponding text triples according to the attribute mapping table and the entity mapping table, and use this result as The final inference result is returned.

Example 1

In this example, multiple cross-ontology knowledge bases are semantically fused by means of semantic linking and fusion. Here we use the cross-ontology massive biomedical knowledge graph as an example. As shown in Figure 2, the graph integrates 20 different knowledge bases , containing nearly 5 billion triples of data. The knowledge map is stored in the HDFS file system in the form of triples (as shown in Table 1) for parallel processing.

Table 1

subject Relation object Subject_text1 Relation_text1 Object_text1 Subject_text2 Relation_text2 Object_text2 Subject_text3 Relation_text3 Object_text3 … … … Subject_textN Relation_textN Object_textN

After constructing the above knowledge graph, a series of inference rules can be obtained through the knowledge graph to express the relationship between entities and entities in a single ontology, and the relationship between entities across ontologies can be expressed through multiple link inference rules, so that by constructing an attribute Chain reasoning network to effectively characterize possible relationships between entities across ontology.

Then rewrite the reasoning network and knowledge map, and use the MapReduce algorithm framework to perform parallel iterative reasoning of attribute chains in accordance with the reasoning network, and modify the corresponding reasoning network for the next round of reasoning. Utilize the method of the present invention to complete the reasoning of herb (Herb) and gene (Gene) association discovery across ontology in biomedicine, the real reasoning result shows that the associated entity pair obtained has very high accuracy, and also has very high calculation operation efficiency.

The above-mentioned specific embodiments have described the technical solutions and beneficial effects of the present invention in detail. It should be understood that the above-mentioned are only the most preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, supplements and equivalent replacements made within the scope shall be included in the protection scope of the present invention.

Claims (3)

1. A cross-ontology attribute chain reasoning method based on cloud platform, comprising the following steps: (1) Use the semantic linking method to fuse ontology data in multiple fields to obtain a knowledge graph H in a unified format; (2) Use the knowledge graph to obtain a series of inference rules A expressing the relationship between entities in a single ontology, and use the Owl2 attribute chain primitive of the semantic ontology description language to model the complex relationship between cross-ontology entities and entities, Obtain multiple attribute chain inference rules B expressing the relationship between entities across ontology entities, and A and B form an attribute chain inference network C; (3) Assign a corresponding attribute id number to each attribute object in the knowledge graph and attribute chain inference network C, and assign a corresponding entity id number to each entity object to form knowledge graph H' and attribute chain inference network C'; (4) Using the MapReduce framework, according to the attribute chain reasoning network C', perform attribute chain parallel reasoning on the knowledge map H', and modify and update the attribute chain reasoning network C'; (5) Save a series of results obtained by reasoning in step (4) into hdfs, and add them to the knowledge map H'; (6) Judging whether the reasoning result of this time is consistent with the reasoning result of the last time, if so, execute step (7); if not, jump to execute step (4); (7) End the reasoning, merge the reasoning results generated by multiple iterations on hdfs, and remove the repeated triples in the reasoning results, and then restore the corresponding text triples according to the attribute mapping table and entity mapping table, and convert the result Returned as the final inference result. 2. the cross-ontology attribute chain reasoning method based on cloud platform as claimed in claim 1, is characterized in that, the concrete process of described step (3) is: Build an attribute mapping table and assign an attribute id number to each attribute object; Build an entity mapping table and assign an entity id number to each entity object; Replace the attribute chain object in the attribute chain reasoning network C with the attribute id to form the attribute chain reasoning network C′; Traverse each triple in the knowledge graph H, and replace the corresponding head node, tail node and relationship node with the entity id and attribute id to form the knowledge graph H'. 3. the cross-ontology attribute chain reasoning method based on cloud platform as claimed in claim 1, is characterized in that, the concrete steps of step (4) are: (4-1) Map stage: take (line number, triplet) key-value pair as input, and output (link attribute id key value, triplet) key-value pair; (4-2) Reduce stage: use the (link attribute id key value, triplet) key-value pair output in the Map stage as the input of this stage, fuse triplets with the same id key value, and output (_, new triplet groups or pending triples); (4-3) Merge and update the adjacent attribute chain objects in the attribute chain reasoning network C′, and reassign a new id for the attribute chain objects; (4-4) Check whether there is a new triple output, if so, jump to step (4-1), if not, output the reasoning result.