Labeling system and method for time-series data

Abstract

A time-series data annotation system includes a memory and a processor. The memory stores multiple instructions. The processor is connected to the memory and accesses these instructions to perform the following operations: dividing a plurality of local changes in the time-series data into a plurality of time segments; extracting features from each of the plurality of time segments to obtain corresponding segment features; performing symbol mapping to map the plurality of segment features into a symbol sequence; generating cue words using a cue word generator based on the symbol sequence and at least one inference example selected from an inference example library based on the symbol sequence; and inputting the cue words into an annotation model to generate annotation results and annotation rationale for the time-series data.

Claims

1. A time-series data annotation system, comprising: a memory for storing a plurality of instructions; and a processor connected to the memory, the processor being configured to perform the following operations when accessing the memory to execute the plurality of instructions: dividing a plurality of local changes in a time-series data into a plurality of time segments; performing feature extraction on each of the plurality of time segments to obtain corresponding segment features; performing a symbol mapping to map the plurality of segment features into a symbol sequence; generating a cue word using a cue word generator based on the symbol sequence and at least one inference example selected from an inference example library based on the symbol sequence; and inputting the cue word into an annotation model to generate an annotation result for the time-series data and an annotation rationale.

2. The time-series data annotation system as described in claim 1, wherein the at least one inference example includes an example symbol, an example label, and an inference reason between the example symbol and the example label.

3. The time-series data annotation system as described in claim 1 further includes: an operating device connected to the processor, the operating device being configured to: display the symbol sequence; and receive at least one suggested annotation corresponding to the symbol sequence to transmit to the processor; and the processor being further configured to update the inference example library based on the at least one suggested annotation.

4. The time series data annotation system as described in claim 1 further includes: an operating device connected to the processor, the operating device being configured to perform the following operations: display the time series data and the corresponding annotation result and the annotation reason; and receive an annotation review of one of the annotation results to transmit to the processor; and the processor being configured to update the inference example library based on the annotation review.

5. The time-series data annotation system as described in claim 4, wherein the processor is further configured to perform the following operations: perform reinforcement learning on the annotation model based on the symbol sequence, the annotation result, the annotation rationale, and the annotation review.

6. The time series data annotation system as described in claim 4, wherein the processor updates the inference example library based on the annotation review to add a new inference example to the inference example library based on the annotation review.

7. The time-series data annotation system as described in claim 6, wherein the annotation model is used to annotate the time-series data of an application domain and generate a confidence level corresponding to the annotation result, and the processor adjusts the prompt word generator to select a retrieval rule for the at least one inference example from the inference example library based on at least one of the novelty of the new inference example, the confidence level, and the domain relevance of the new inference example to the application domain.

8. The time-series data annotation system as described in claim 1, wherein the prompt word generator selects the at least one inference example from the inference example library and generates the prompt word using retrieval enhancement generation.

9. A time series data annotation method, performed by a processor, comprising: dividing a plurality of local variations in a time series data into a plurality of time segments; performing feature extraction on each of the plurality of time segments to obtain corresponding segment features; performing a symbol mapping to map the plurality of segment features into a symbol sequence; generating a cue word using a cue word generator based on the symbol sequence and at least one inference example selected from an inference example library based on the symbol sequence; and inputting the cue word into an annotation model to generate an annotation result for the time series data and an annotation rationale.

10. The time-series data annotation method as described in claim 9 further includes: receiving annotation review of one of the annotation results; and updating the inference example library based on the annotation review.

Images