Automatically converting error logs having different format types into a standardized and labeled format having relevant natural language information
Embodiments of the invention are directed to computer-implemented methods of labeling unlabeled electronic information. In a non-limiting embodiment of the invention, the computer-implemented method includes receiving, using a processor system, an unlabeled error log (EL) having an EL format. A set of unlabeled EL keywords are extracted from the unlabeled EL. For each unlabeled EL keyword, the processor system uses the unlabeled EL keyword to extract an electronic document (ED) sentence from an ED based on a determination that the ED sentence is relevant to the unlabeled EL keyword. One or more ED keywords are extracted from the ED sentence. A deep neural network of the processor system is used to predict an ED sentence label for the ED sentence, an ED keyword label for the ED keyword, and an EL keyword label for the unlabeled EL keyword.
1 . A computer-implemented method of executing a predictive labeling pipeline of a processor system, the computer-implemented method comprising:
receiving, using the processor system and an input communications path, unlabeled error logs (ELs) during a runtime of software from which the unlabeled ELs are generated, the software comprising systems software and database management (DBM) software, the systems software configured to manage software operations of a computer that hosts the systems software, the DBM software configured to control data storage, retrieval and organization operations performed by a database;
wherein EL formats of the unlabeled ELs comprise:
systems software machine-generated codes created for communication between non-human electronic entities;
DBM software machine-generated codes created for communication between non-human electronic entities; and
text not in a sentence format;
wherein the predictive labeling pipeline comprises a plurality of interconnected modules including a keyword extraction module, a context-based search and prediction module, and a neural network prediction module, each implemented as executable code stored in memory and executed by the processor system;
placing, using the processor system, the unlabeled ELs into the predictive labeling pipeline; and
using the plurality of interconnected modules to apply, during the runtime of the software from which the unlabeled ELs are generated, multiple iterations of a set of pipeline operations to the predictive labeling pipeline, wherein the set of pipeline operations converts each of the unlabeled ELs to a labeled EL having a structured EL format;
wherein the set of pipeline operations include context-based search and prediction operations;
wherein the context-based search and prediction operations comprise, responsive to one or more keywords extracted using the keyword extraction module from one or more of the unlabeled ELs, using the context-based search and prediction module to search a plurality of electronic documents (EDs) in a plurality of locations and apply a keyphrase extraction and prediction algorithm to the plurality of EDs in the plurality of locations to select and extract one or more context-aware natural language (NL) description sentences that the keyphrase extraction and prediction algorithm predicts to provide context and descriptions of issues associated with one or more of the unlabeled ELs:
wherein the search-based pipeline operations further comprise using the context-based search and prediction module to apply a keyphrase extraction and prediction algorithm to the one or more context-aware NL description sentences to extract NL description sentence keywords from the one or more context-aware NL description sentences; and
wherein the set of pipeline operations converts each of the unlabeled ELs to an instance of the labeled EL having the structured EL format based at least in part on the one or more keywords extracted from one or more of the unlabeled ELs, the context-aware NL description sentences, and the one or more NL description sentence keywords extracted from the context-aware NL description sentences.
2 . The computer-implemented method of claim 1 , wherein:
the set of pipeline operations include label prediction pipeline operations; and
the label prediction pipeline operations comprise using the neural network prediction module to apply a NL-based question and answer (Q&A) prediction algorithm to the one or more keywords extracted from one or more of the unlabeled ELs, the context-aware NL description sentences, and the one or more NL description sentence keywords extracted from the context-aware NL description sentences to generate predicted labels for the one or more keywords extracted from one or more of the unlabeled ELs, the context-aware NL description sentences, and the one or more NL description sentence keywords extracted from the context-aware NL description sentences.
3 . The computer-implemented method of claim 2 , wherein the NL-based Q&A prediction algorithm is trained using a training data set that includes multiple types of ELs that have been labeled at a level of detail that applies to multiple types of the software from which the unlabeled ELs are generated.
4 . The computer-implemented method of claim 2 , wherein the predicted labels comprise metadata that provides predicted answer information about an underlying performance issue that caused the software to generate an associated one of the unlabeled ELs.
5 . The computer-implemented method of claim 4 , wherein the set of pipeline operations converts each of the unlabeled ELs to an instance of the labeled EL having the structured EL format further based at least in part on the predicted labels.
6 . The computer-implemented method of claim 4 , wherein the set of pipeline operations converts each of the unlabeled ELs to an instance of the labeled EL having the structured EL format further based at least in part on populating user-selected fields of the structured EL format with the predicted labels.
7 . The computer-implemented method of claim 1 , wherein:
the context-based search and prediction module is configured to access the plurality of locations through one or more networks; and
the plurality of EDs is selected from the group consisting of a website posting, an article, and an operating manual.
8 . The computer-implemented method of claim 1 further comprising controlling, by a post-processing module, a topic and entity relation extraction module to use the labeled EL having the structured EL format as a guide to electronically search for and locate relevant EDs from which topic and entity relation information can be extracted.
9 . A computer system comprising:
a memory; and
a processor system communicatively coupled to the memory, wherein the processor system is configured to perform processor system operations that execute a predictive labeling pipeline, the processor system operations comprising:
receiving using the processor system and an input communications path unlabeled error logs (ELs) during a runtime of software from which the unlabeled ELs are generated, the software comprising systems software and database management (DBM) software, the systems software configured to manage software operations of a computer that hosts the systems software, the DBM software configured to control data storage, retrieval and organization operations performed by a database;
wherein EL formats of the unlabeled ELs comprise:
systems software machine-generated codes created for communication between non-human electronic entities;
DBM software machine-generated codes created for communication between non-human electronic entities; and
text not in a sentence format;
wherein the predictive labeling pipeline comprises a plurality of interconnected modules including a keyword extraction module, a context-based search and prediction module, and a neural network prediction module, each implemented as executable code stored in memory and executed by the processor system;
placing, using the processor system, the unlabeled ELs into the predictive labeling pipeline; and
using the plurality of interconnected modules to apply, during the runtime of the software from which the unlabeled ELs are generated, multiple iterations of a set of pipeline operations to the predictive labeling pipeline, wherein the set of pipeline operations converts each of the unlabeled ELs to a labeled EL having a structured EL format;
wherein the set of pipeline operations include context-based search and prediction operations;
wherein the context-based search and prediction operations comprise, responsive to one or more keywords extracted using the keyword extraction module from one or more of the unlabeled ELs, using the context-based search and prediction module to search a plurality of electronic documents (EDs) in a plurality of locations and apply a keyphrase extraction and prediction algorithm to the plurality of EDs in the plurality of locations to select and extract one or more context-aware natural language (NL) description sentences that the keyphrase extraction and prediction algorithm predicts to provide context and descriptions of issues associated with one or more of the unlabeled ELs:
wherein the search-based pipeline operations further comprise using the context-based search and prediction module to apply a keyphrase extraction and prediction algorithm to the one or more context-aware NL description sentences to extract NL description sentence keywords from the one or more context-aware NL description sentences; and
wherein the set of pipeline operations converts each of the unlabeled ELs to an instance of the labeled EL having the structured EL format based at least in part on the one or more keywords extracted from one or more of the unlabeled ELs, the context-aware NL description sentences, and the one or more NL description sentence keywords extracted from the context-aware NL description sentences.
10 . The computer system of claim 9 , wherein:
the set of pipeline operations include label prediction pipeline operations; and
the label prediction pipeline operations comprise using the neural network prediction module to apply a NL-based question and answer (Q&A) prediction algorithm to the one or more keywords extracted from one or more of the unlabeled ELs, the context-aware NL description sentences, and the one or more NL description sentence keywords extracted from the context-aware NL description sentences to generate predicted labels for the one or more keywords extracted from one or more of the unlabeled ELs, the context-aware NL description sentences, and the one or more NL description sentence keywords extracted from the context-aware NL description sentences.
11 . The computer system of claim 10 , wherein the NL-based Q&A prediction algorithm is trained using a training data set that includes multiple types of ELs that have been labeled at a level of detail that applies to multiple types of the software from which the unlabeled ELs are generated.
12 . The computer system of claim 10 , wherein the predicted labels comprise metadata that provides predicted answer information about an underlying performance issue that caused the software to generate an associated one of the unlabeled ELs.
13 . The computer system of claim 12 , wherein the set of pipeline operations converts each of the unlabeled ELs to an instance of the labeled EL having the structured EL format further based at least in part on the predicted labels.
14 . The computer system of claim 12 , wherein the set of pipeline operations converts each of the unlabeled ELs to an instance of the labeled EL having the structured EL format further based at least in part on populating user-selected fields of the structured EL format with the predicted labels.
15 . The computer system of claim 9 , wherein:
the context-based search and prediction module is configured to access the plurality of locations through one or more networks; and
the plurality of EDs is selected from the group consisting of a website posting, an article, and an operating manual.
16 . The computer system of claim 9 , wherein the processor system operations further comprise controlling, by a post-processing module, a topic and entity relation extraction module to use the labeled EL having the structured EL format as a guide to electronically search for and locate relevant EDs from which topic and entity relation information can be extracted.
17 . A computer program product comprising a computer readable program stored on a computer readable storage medium, wherein the computer readable program when executed by a processor system causes the processor system to perform processor operations that execute a predictive labeling pipeline, the processor system operations comprising:
receiving using the processor system and an input communications path unlabeled error logs (ELs) during a runtime of software from which the unlabeled ELs are generated, the software comprising systems software and database management (DBM) software, the systems software configured to manage software operations of a computer that hosts the systems software, the DBM software configured to control data storage, retrieval and organization operations performed by a database;
wherein EL formats of the unlabeled ELs comprise: EL having the EL format comprises:
systems software machine-generated codes created for communication between non-human electronic entities;
DBM software machine-generated codes created for communication between non-human electronic entities; and
text not in a sentence format;
wherein the predictive labeling pipeline comprises a plurality of interconnected modules including a keyword extraction module, a context-based search and prediction module, and a neural network prediction module, each implemented as executable code stored in memory and executed by the processor system;
placing, using the processor system, the unlabeled ELs into the predictive labeling pipeline; and
using the plurality of interconnected modules to apply, during the runtime of the software from which the unlabeled ELs are generated, multiple iterations of a set of pipeline operations to the predictive labeling pipeline, wherein the set of pipeline operations converts each of the unlabeled ELs to a labeled EL having a structured EL format;
wherein the set of pipeline operations include context-based search and prediction operations;
wherein the context-based search and prediction operations comprise, responsive to one or more keywords extracted using the keyword extraction module from one or more of the unlabeled ELs, using the context-based search and prediction module to search a plurality of electronic documents (EDs) in a plurality of locations and apply a keyphrase extraction and prediction algorithm to the plurality of EDs in the plurality of locations to select and extract one or more context-aware natural language (NL) description sentences that the keyphrase extraction and prediction algorithm predicts to provide context and descriptions of issues associated with one or more of the unlabeled ELs:
wherein the search-based pipeline operations further comprise using the context-based search and prediction module to apply a keyphrase extraction and prediction algorithm to the one or more context-aware NL description sentences to extract NL description sentence keywords from the one or more context-aware NL description sentences; and
wherein the set of pipeline operations converts each of the unlabeled ELs to an instance of the labeled EL having the structured EL format based at least in part on the one or more keywords extracted from one or more of the unlabeled ELs, the context-aware NL description sentences, and the one or more NL description sentence keywords extracted from the context-aware NL description sentences.
18 . The computer program product of claim 17 , wherein:
the set of pipeline operations include label prediction pipeline operations; and
the label prediction pipeline operations comprise using the neural network prediction module to apply a NL-based question and answer (Q&A) prediction algorithm to the one or more keywords extracted from one or more of the unlabeled ELs, the context-aware NL description sentences, and the one or more NL description sentence keywords extracted from the context-aware NL description sentences to generate predicted labels for the one or more keywords extracted from one or more of the unlabeled ELs, the context-aware NL description sentences, and the one or more NL description sentence keywords extracted from the context-aware NL description sentences.
19 . The computer program product of claim 18 , wherein the NL-based Q&A prediction algorithm is trained using a training data set that includes multiple types of ELs that have been labeled at a level of detail that applies to multiple types of the software from which the unlabeled ELs are generated.
20 . The computer program product of claim 17 , wherein the processor system operations further comprise controlling, by a post-processing module, a topic and entity relation extraction module to use the labeled EL having the structured EL format as a guide to electronically search for and locate relevant EDs from which topic and entity relation information can be extracted.