Systems and methods for automatically classifying a fault report related to a mechanical asset
A plurality of object names and a query string are defined. A first set that includes a plurality of first query string-object name pairs is generated based on the plurality of object names and the query string. A closeness score is determined for each first query string-object name pair in the first set. A second set that includes a plurality of second query string-object name pairs is generated, wherein the second query string-object name pairs are selected from the first set based on the closeness scores. A measure of semantic similarity is determined for each second query string-object name pair in the second set. A third set of third query string-object name pairs is defined, wherein the third query string-object name pairs are selected from the second set based on the measures of semantic similarity.
1. A method comprising:
defining a plurality of object names;
defining a query string;
generating a first set that includes a plurality of first query string-object name pairs based on the plurality of object names and the query string;
determining a closeness score for each first query string-object name pair in the first set, wherein determining the closeness score comprises: applying to each query string-object name pair in the first set an approximate string matching technique to generate a respective closeness score for the respective query string-object name pair;
defining a second set that includes a plurality of second query string-object name pairs, the second query string-object name pairs selected from the first set based on the closeness scores determined, wherein defining the second set comprises: including in the second set all query string-object name pairs having a closeness score that equals or exceeds a predetermined level;
determining a measure of semantic similarity for each second query string-object name pair in the second set, wherein determining the measure of semantic similarity comprises: applying to each query string-object name pair in the second set a semantic matching language model to generate a measure of semantic similarity for the respective query string-object name pair; and
defining a third set of third query string-object name pairs, the third query string-object name pairs selected from the second set based on the measures of semantic similarity determined, wherein defining the third set of third query string-object name pairs comprises: ranking the query string-object name pairs in the second set based on measures of semantic similarity; and
selecting a number of highest-ranked query string-object name pairs.
2. The method of claim 1 , wherein each object name is a name of a component of a mechanical asset.
3. The method of claim 1 , wherein defining a plurality of object names further comprises:
receiving a list of assets comprising the plurality of asset names organized in a tree structure; and
modifying the list of assets to generate a list comprising the plurality of object names without the tree structure.
4. The method of claim 1 , wherein defining a query string further comprises:
receiving a string of text from a user; and
cleaning the string of text.
5. The method of claim 1 , wherein generating a first set that includes a plurality of first query string-object name pairs further comprises:
generating a first set comprising all possible query string-object name combinations based on the plurality of object names and the query string.
6. The method of claim 1 , wherein the semantic matching language model includes a language model based on the open-source Bidirectional Encoder Representations from Transformers (BERT) architecture published by Google AI.
7. A non-transitory computer readable medium having stored thereon software instructions that, when executed by a processor, cause the processor to perform a series of operations comprising:
defining a plurality of object names;
defining a query string;
generating a first set that includes a plurality of first query string-object name pairs based on the plurality of object names and the query string;
determining a closeness score for each first query string-object name pair in the first set, wherein determining the closeness score comprises: applying to each query string-object name pair in the first set an approximate string matching technique to generate a respective closeness score for the respective query string-object name pair;
defining a second set that includes a plurality of second query string-object name pairs, the second query string-object name pairs selected from the first set based on the closeness scores determined, wherein defining the second set comprises: including in the second set all query string-object name pairs having a closeness score that equals or exceeds a predetermined level;
determining a measure of semantic similarity for each second query string-object name pair in the second set, wherein determining the measure of semantic similarity comprises: applying to each query string-object name pair in the second set a semantic matching language model to generate a measure of semantic similarity for the respective query string-object name pair; and
defining a third set of third query string-object name pairs, the third query string-object name pairs selected from the second set based on the measures of semantic similarity determined, wherein defining the third set of third query string-object name pairs comprises: ranking the query string-object name pairs in the second set based on measures of semantic similarity; and
selecting a number of highest-ranked query string-object name pairs.
8. The non-transitory computer readable medium of claim 7 , wherein each object name is a name of a component of a mechanical asset.
9. The non-transitory computer readable medium of claim 7 , wherein defining a plurality of object names further comprises:
receiving a list of assets comprising the plurality of asset names organized in a tree structure; and
modifying the list of assets to generate a list comprising the plurality of object names without the tree structure.
10. The non-transitory computer readable medium of claim 7 , wherein defining a query string further comprises:
receiving a string of text from a user; and
cleaning the string of text.
11. The non-transitory computer readable medium of claim 7 , wherein generating a first set that includes a plurality of first query string-object name pairs further comprises:
generating a first set comprising all possible query string-object name combinations based on the plurality of object names and the query string.
12. The non-transitory computer readable medium of claim 7 , wherein the semantic matching language model includes a language model based on the open-source Bidirectional Encoder Representations from Transformers (BERT) architecture published by Google AI.
13. A system comprising one or more processors, and a non-transitory computer-readable medium including one or more sequences of instructions that, when executed by the one or more processors, cause the system to perform operations comprising:
defining a plurality of object names;
defining a query string;
generating a first set that includes a plurality of first query string-object name pairs based on the plurality of object names and the query string;
determining a closeness score for each first query string-object name pair in the first set, wherein determining the closeness score comprises: applying to each query string-object name pair in the first set an approximate string matching technique to generate a respective closeness score for the respective query string-object name pair;
defining a second set that includes a plurality of second query string-object name pairs, the second query string-object name pairs selected from the first set based on the closeness scores determined, wherein defining the second set comprises: including in the second set all query string-object name pairs having a closeness score that equals or exceeds a predetermined level;
determining a measure of semantic similarity for each second query string-object name pair in the second set, wherein determining the measure of semantic similarity comprises: applying to each query string-object name pair in the second set a semantic matching language model to generate a measure of semantic similarity for the respective query string-object name pair; and
defining a third set of third query string-object name pairs, the third query string-object name pairs selected from the second set based on the measures of semantic similarity determined, wherein defining the third set of third query string-object name pairs comprises: ranking the query string-object name pairs in the second set based on measures of semantic similarity; and
selecting a number of highest-ranked query string-object name pairs.