Training method and apparatus for fault recognition model, fault recognition method and apparatus, and electronic device
View Patent ↗Provided are a training method and apparatus for a fault recognition model, a fault recognition method and apparatus, an electronic device, and a medium. A training method for a fault recognition model includes: obtaining a first fault data set; generating a first sample set based on the first fault data set, where each sample in the first sample set includes a fault description text, a corresponding faulty part, and a corresponding fault type; and adjusting parameters of a fault recognition model by using the fault description text of the sample in the first sample set as an input to the fault recognition model, such that the fault recognition model outputs a first label and a second label, where the first label is the faulty part corresponding to the input fault description text, and the second label is the fault type corresponding to the faulty part.
1 . A computer-implemented training method for a fault recognition model, comprising:
obtaining a first fault data set, wherein each piece of fault data in the first fault data set comprises at least a fault phenomenon and a fault recognition result;
generating a first sample set based on the first fault data set, wherein each sample in the first sample set comprises a fault description text, a corresponding faulty part, and a corresponding fault type;
outputting, by a fault recognition model and by using the fault description text of the sample in the first sample set as an input, a two-level label structure comprising a first label as a parent label and a second label as a child label, wherein the first label is the faulty part corresponding to input fault description text, and wherein the second label is the fault type corresponding to the faulty part; and
adjusting parameters of the fault recognition model by using at least a first loss function and a second loss function, wherein a difference between the parent label of the two-level label structure and a real label is identified based on the first loss function, and wherein a difference between the parent label and the child label of the two-level label structure is identified based on the second loss function.
2 . The method according to claim 1 , wherein the first loss function is a cross entropy loss function, and wherein the second loss function is a mean square error loss function.
3 . The method according to claim 1 , wherein generating a first sample set based on the first fault data set comprises:
determining a plurality of faulty parts and a fault type corresponding to each of the plurality of faulty parts by processing the fault recognition result in the fault data of the first fault data set; and
generating a first sample set by associating the fault description text, the corresponding faulty part, and the corresponding fault type in the fault data of the first fault data set as a sample.
4 . The method according to claim 3 , wherein generating a first sample set based on the first fault data set further comprises:
providing, in response to a number of samples having a specific faulty part and a specific fault type being less than a threshold, additional associated fault description texts for the specific faulty part and the specific fault type by at least one of the following operations: synonym replacement, Chinese-English translation, word sequence exchanging, and word deletion.
5 . The method according to claim 1 , further comprising:
after adjusting parameters of the fault recognition model, obtaining a second fault data set, wherein the second fault data set is different from the first fault data set;
generating a second sample set based on the second fault data set; and
further adjusting the parameters of the fault recognition model by using fault description texts of at least one sample in the first sample set and at least one sample in the second sample set as inputs to the fault recognition model.
6 . A fault recognition method, comprising:
obtaining a first text, wherein the first text is a text for describing a fault phenomenon; and
inputting the first text to the fault recognition model trained by using the method according to claim 1 , to obtain an output result,
wherein the output result comprises a faulty part and a fault type corresponding to the faulty part.
7 . An electronic device, comprising:
at least one processor; and
a memory communicatively connected to the at least one processor, wherein
the memory stores instructions executable by the at least one processor, and when executed by the at least one processor, the instructions cause the at least one processor to perform processing comprising:
obtaining a first fault data set, wherein each piece of fault data in the first fault data set comprises at least a fault phenomenon and a fault recognition result;
generating a first sample set based on the first fault data set, wherein each sample in the first sample set comprises a fault description text, a corresponding faulty part, and a corresponding fault type; and
outputting, by a fault recognition model and by using the fault description text of the sample in the first sample set as an input, a two-level label structure comprising a first label as a parent label and a second label as a child label, wherein the first label is the faulty part corresponding to input fault description text, and wherein the second label is the fault type corresponding to the faulty part,
wherein the instructions further cause the at least one processor to perform processing comprising adjusting parameters of the fault recognition model by using at least a first loss function and a second loss function, wherein a difference between the parent label of the two-level label structure and a real label is identified based on the first loss function, and wherein a difference between the parent label and the child label of the two-level label structure is identified based on the second loss function.
8 . The electronic device according to claim 7 , wherein the first loss function is a cross entropy loss function, and wherein the second loss function is a mean square error loss function.
9 . The electronic device according to claim 7 , wherein generating a first sample set based on the first fault data set comprises:
determining a plurality of faulty parts and a fault type corresponding to each of the plurality of faulty parts by processing the fault recognition result in the fault data of the first fault data set; and
generating a first sample set by associating the fault description text, the corresponding faulty part, and the corresponding fault type in the fault data of the first fault data set as a sample.
10 . The electronic device according to claim 9 , wherein generating a first sample set based on the first fault data set further comprises:
providing, in response to a number of samples having a specific faulty part and a specific fault type being less than a threshold, additional associated fault description texts for the specific faulty part and the specific fault type by at least one of the following operations: synonym replacement, Chinese-English translation, word sequence exchanging, and word deletion.
11 . The electronic device according to claim 7 , wherein the instructions further cause the at least one processor to perform processing further comprising:
after adjusting parameters of the fault recognition model, obtaining a second fault data set, wherein the second fault data set is different from the first fault data set;
generating a second sample set based on the second fault data set; and
further adjusting the parameters of the fault recognition model by using fault description texts of at least one sample in the first sample set and at least one sample in the second sample set as inputs to the fault recognition model.
12 . A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to perform processing comprising:
obtaining a first fault data set, wherein each piece of fault data in the first fault data set comprises at least a fault phenomenon and a fault recognition result;
generating a first sample set based on the first fault data set, wherein each sample in the first sample set comprises a fault description text, a corresponding faulty part, and a corresponding fault type; and
outputting, by a fault recognition model and by using the fault description text of the sample in the first sample set as an input, a two-level label structure comprising a first label as a parent label and a second label as a child label, wherein the first label is the faulty part corresponding to input fault description text, and wherein the second label is the fault type corresponding to the faulty part,
wherein the computer instructions are further to cause the computer to perform processing comprising adjusting parameters of the fault recognition model by using at least a first loss function and a second loss function, wherein a difference between the parent label of the two-level label structure and a real label is identified based on the first loss function, and wherein a difference between the parent label and the child label of the two-level label structure is identified based on the second loss function.
13 . The non-transitory computer-readable storage medium according to claim 12 , wherein the first loss function is a cross entropy loss function, and wherein the second loss function is a mean square error loss function.
14 . The non-transitory computer-readable storage medium according to claim 12 , wherein generating a first sample set based on the first fault data set comprises:
determining a plurality of faulty parts and a fault type corresponding to each of the plurality of faulty parts by processing the fault recognition result in the fault data of the first fault data set; and
generating a first sample set by associating the fault description text, the corresponding faulty part, and the corresponding fault type in the fault data of the first fault data set as a sample.
15 . The non-transitory computer-readable storage medium according to claim 14 , wherein generating a first sample set based on the first fault data set further comprises:
providing, in response to a number of samples having a specific faulty part and a specific fault type being less than a threshold, additional associated fault description texts for the specific faulty part and the specific fault type by at least one of the following operations: synonym replacement, Chinese-English translation, word sequence exchanging, and word deletion.
16 . The non-transitory computer-readable storage medium according to claim 12 , wherein the instructions further cause at least one processor to perform processing further comprising:
after adjusting parameters of the fault recognition model, obtaining a second fault data set, wherein the second fault data set is different from the first fault data set;
generating a second sample set based on the second fault data set; and
further adjusting the parameters of the fault recognition model by using fault description texts of at least one sample in the first sample set and at least one sample in the second sample set as inputs to the fault recognition model.