Autonomous system for hypotheses generation

1. A method for evaluating relatedness of a pair of independent concepts by a bioinformatics system, the method comprising:

receiving a query input comprising a pair of independent concepts having no formally recognized relationship in a knowledgebase of a domain of knowledge, wherein the knowledgebase comprises an bipartite graph having a set of concepts having known interactions, and a set of electronic documents, wherein a given pair of related concepts is connected to an electronic document where the related concepts co-occur in the electronic document, and wherein concepts of the query input and concepts of the set of electronic documents are detected using a concept-detection tool performing natural language processing, the concept detection tool having a user interface to receive the query input;

identifying a plurality of candidate connecting concepts for the pair of independent concepts by analyzing a plurality of electronic documents, a plurality of co-occurrence matrices organized as a time series, and a plurality of lists of similar concepts detected using natural language processing, wherein the identifying further comprises:

selecting a plurality of candidate connecting concepts meeting a threshold

selection value of a distance measure of that concept to the independent concept as reflected in the knowledgebase;

generating a plurality of hypotheses for the pair of independent concepts based on the identified candidate connecting concepts, wherein the plurality of hypotheses each represent a proposition that the first and second independent concepts are related;

ranking the plurality of hypotheses relative to one another, the ranking comprising:

traversing vertices in the knowledgebase to identify a plurality of paths connecting the first independent concept to the second independent concept; and

applying a classifier model trained using features of the selected plurality of candidate connecting concepts, a temporal prediction model trained using temporal properties of the plurality of input documents, and a word embedding evolution model trained using a plurality of word embeddings organized as a time series, to the plurality of paths, to generate rankings of the plurality of hypotheses;

selecting top-k ranked hypotheses from among the plurality of hypotheses; and

displaying results of the selecting to a user in response to receiving the query.

2. The method of claim 1 , wherein identifying a plurality of candidate connecting concepts further comprises:

training the classifier model using the features of the selected plurality of candidate connecting concepts.

3. The method of claim 2 , further comprising:

training the temporal prediction model using the temporal properties of the plurality of input documents.

4. The method of claim 3 , further comprising:

training the word embedding evolution model using the plurality of word embeddings organized as a time series.

5. The method of claim 1 , further comprising:

processing an additional pair of independent concepts by performing one or more steps performed with respect to the pair of independent concepts;

generating, for each of the pair of independent concepts and for the additional pair of independent concepts, one or more hypothesis worlds; and

ranking hypothesis worlds according to a measure of their corresponding independent concepts drifting closer to one another.

6. A computer program product for evaluating relatedness of a pair of independent concepts by a bioinformatics system, the computer program product comprising a non-transitory tangible storage device having program code embodied therewith, the program code executable by a processor of a computer to perform a method, the method comprising:

receiving, by the processor, a query input comprising a pair of independent concepts having no formally recognized relationship in a knowledgebase of a domain of knowledge, wherein the knowledgebase comprises an bipartite graph having a set of concepts having known interactions, and a set of electronic documents, wherein a given pair of related concepts is connected to an electronic document where the related concepts co-occur in the electronic document, and wherein concepts of the query input and concepts of the set of electronic documents are detected using a concept-detection tool performing natural language processing, the concept detection tool having a user interface to receive the query input;

identifying, by the processor, a plurality of candidate connecting concepts for the pair of independent concepts by analyzing a plurality of electronic documents, a plurality of co-occurrence matrices organized as a time series, and a plurality of lists of similar concepts detected using natural language wherein the identifying further comprises:

selecting a plurality of candidate connecting concepts meeting a threshold

selection value of a distance measure of that concept to the independent concept as reflected in the knowledgebase;

generating, by the processor, a plurality of hypotheses for the pair of independent concepts based on the identified candidate connecting concepts, wherein the plurality of hypotheses each represent a proposition that the first and second independent concepts are related;

ranking, by the processor, the plurality of hypotheses relative to one another, the ranking comprising:

traversing, by the processor, vertices in the knowledgebase to identify a plurality of paths connecting the first independent concept to the second independent concept; and

applying, by the processor, a classifier model trained using features of the selected plurality of candidate connecting concepts, a temporal prediction model trained using temporal properties of the plurality of input documents, and a word embedding evolution model trained using a plurality of word embeddings organized as a time series, to the plurality of paths, to generate rankings of the plurality of hypotheses;

selecting, by the processor, top-k ranked hypotheses from among the plurality of hypotheses; and

displaying, by the processor, results of the selecting to a user in response to receiving the query.

7. The computer program product of claim 6 , wherein identifying a plurality of candidate connecting concepts further comprises:

training, by the processor, the classifier model using the features of the selected plurality of candidate connecting concepts.

8. The computer program product of claim 7 , wherein the performed method further comprises:

training, by the processor, the temporal prediction model using the temporal properties of the plurality of input documents.

9. The computer program product of claim 8 , wherein the performed method further comprises:

training the word embedding evolution model using the plurality of word embeddings organized as a time series.

10. The computer program product of claim 7 , wherein the performed method further comprises:

processing, by the processor, an additional pair of independent concepts by performing one or more steps performed with respect to the pair of independent concepts;

generating, by the processor, for each of the pair of independent concepts and for the additional pair of independent concepts, one or more hypothesis worlds; and

ranking, by the processor, hypothesis worlds according to a measure of their corresponding independent concepts drifting closer to one another.

11. A computer system for evaluating relatedness of a pair of independent concepts by a bioinformatics system, comprising:

one or more computer devices each having one or more processors and one or more tangible storage devices; and

a program embodied on at least one of the one or more storage devices, the program having a plurality of program instructions for execution by the one or more processors, the program instructions comprising instructions for:

receiving a query input comprising a pair of independent concepts having no formally recognized relationship in a knowledgebase of a domain of knowledge, wherein the knowledgebase comprises an bipartite graph having a set of concepts having known interactions, and a set of electronic documents, wherein a given pair of related concepts is connected to an electronic document where the related concepts co-occur in the electronic document, and wherein concepts of the query input and concepts of the set of electronic documents are detected using a concept-detection tool performing natural language processing, the concept detection tool having a user interface to receive the query input;

identifying a plurality of candidate connecting concepts for the pair of independent concepts by analyzing a plurality of electronic documents, a plurality of co-occurrence matrices organized as a time series, and a plurality of lists of similar concepts detected using natural language processing, wherein the identifying further comprises:

selecting a plurality of candidate connecting concepts meeting a threshold

selection value of a distance measure of that concept to the independent concept as reflected in the knowledgebase;

generating a plurality of hypotheses for the pair of independent concepts based on the identified candidate connecting concepts, wherein the plurality of hypotheses each represent a proposition that the first and second independent concepts are related;

ranking the plurality of hypotheses relative to one another, the ranking comprising:

traversing vertices in the knowledgebase to identify a plurality of paths connecting the first independent concept to the second independent concept; and

applying a classifier model trained using features of the selected plurality of candidate connecting concepts, a temporal prediction model trained using temporal properties of the plurality of input documents, and a word embedding evolution model trained using a plurality of word embeddings organized as a time series, to the plurality of paths, to generate rankings of the plurality of hypotheses;

selecting top-k ranked hypotheses from among the plurality of hypotheses; and

displaying results of the selecting to a user in response to receiving the query.

12. The computer system of claim 11 , wherein identifying a plurality of candidate connecting concepts further comprises:

training the classifier model using the features of the selected plurality of candidate connecting concepts;

training, by the processor, the temporal prediction model using the temporal properties of the plurality of input documents; and

training the word embedding evolution model using the plurality of word embeddings organized as a time series.

13. The computer system of claim 11 , wherein the instructions further comprise instructions for:

processing an additional pair of independent concepts by performing one or more steps performed with respect to the pair of independent concepts;

generating for each of the pair of independent concepts and for the additional pair of independent concepts, one or more hypothesis worlds; and

ranking hypothesis worlds according to a measure of their corresponding independent concepts drifting closer to one another.