Intelligent task navigation system and method based on machine learning

Abstract

The present application discloses an intelligent task navigation system and method based on machine learning, which establishes an interactive relationship with a user through voice, then performs voice recognition on the task description voice input by the user to convert it into text data, and performs semantic understanding and analysis on the text data containing user intention information through natural language processing technology based on machine learning, and further learns and recognizes the user's deep semantic meaning through commonality extraction of semantic features, thereby determining the type of task the user wants to perform, and further matching it with the tasks in the task database.

Description

Technical Field

[0001] The present application relates to the field of intelligent navigation, and more specifically, to an intelligent task navigation system and method based on machine learning. Background Art

[0002] Task navigation refers to the process of guiding users through a series of steps or options to complete a specific task in a computer system or software application. However, as the application of task navigation becomes more and more widespread, simply relying on fixed commands or preset options may not truly understand the user's intention, which limits the flexibility of task navigation and reduces the user experience.

[0003] With the rapid development of artificial intelligence technology, machine learning, as one of its core branches, has demonstrated its strong application potential in many fields and provided new solutions for optimizing task navigation. Summary of the invention

[0004] In order to solve the above technical problems, this application is proposed. The embodiment of this application provides an intelligent task navigation system and method based on machine learning, which establishes an interactive relationship with the user through voice, then performs voice recognition on the task description voice input by the user to convert it into text data, and performs semantic understanding and analysis on the text data containing user intention information through natural language processing technology based on machine learning, and further extracts the commonality of semantic features to learn and identify the deep semantic meaning of the user, thereby determining the type of task the user wants to perform, and further matching it with the tasks in the task database.

[0005] According to one aspect of the present application, there is provided a method for intelligent task navigation based on machine learning, which comprises:

[0006] Obtaining the task description voice input by the user;

[0007] Performing speech recognition on the task description speech to obtain a task description speech recognition result;

[0008] Performing text semantic understanding analysis on the task description speech recognition result to obtain a set of task description semantic encoding feature vectors;

[0009] Passing the set of task description semantic encoding feature vectors through a common semantic feature extraction network based on adaptive gating logic to obtain a task description common extraction semantic encoding feature vector;

[0010] Extracting semantic encoding feature vectors based on commonalities of the task descriptions to determine intent recognition results;

[0011] A task matching the intention recognition result is extracted from a task database.

[0012] According to another aspect of the present application, there is provided an intelligent task navigation system based on machine learning, comprising:

[0013] An information acquisition module, used to acquire the task description voice input by the user;

[0014] A speech recognition module, used for performing speech recognition on the task description speech to obtain a task description speech recognition result;

[0015] A text semantic understanding and analysis module, used for performing text semantic understanding and analysis on the task description speech recognition result to obtain a set of task description semantic encoding feature vectors;

[0016] A common semantic feature extraction module, used for passing the set of task description semantic encoding feature vectors through a common semantic feature extraction network based on adaptive gating logic to obtain a task description common extraction semantic encoding feature vector;

[0017] A recognition result determination module, used to extract semantic encoding feature vectors based on the commonality of the task description to determine the intention recognition result;

[0018] The task matching module is used to extract tasks matching the intention recognition result from the task database.

[0019] Compared with the prior art, the present application provides an intelligent task navigation system and method based on machine learning, which establish an interactive relationship with the user through voice, then perform voice recognition on the task description voice input by the user to convert it into text data, and perform semantic understanding and analysis on the text data containing user intention information through natural language processing technology based on machine learning, and further learn and identify the user's deep semantic meaning through commonality extraction of semantic features, so as to determine the type of task the user wants to perform, and further match it with the tasks in the task database. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0021] Figure 1 is a flowchart of an intelligent task navigation method based on machine learning according to an embodiment of the present application;

[0022] Figure 2is a system architecture diagram of an intelligent task navigation method based on machine learning according to an embodiment of the present application;

[0023] Figure 3 is a flow chart of the training phase of the intelligent task navigation method based on machine learning according to an embodiment of the present application;

[0024] Figure 4 is a flowchart of sub-step S3 of the intelligent task navigation method based on machine learning according to an embodiment of the present application;

[0025] Figure 5 4 is a block diagram of an intelligent task navigation system based on machine learning according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0027] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0028] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0029] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0030] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0031] With the rapid development of artificial intelligence technology, machine learning, as one of its core branches, has demonstrated its strong application potential in many fields and provided new solutions for optimizing task navigation.

[0032] In the technical solution of the present application, a method for intelligent task navigation based on machine learning is proposed. Figure 1 Flow chart of an intelligent task navigation method based on machine learning according to an embodiment of the present application. Figure 2 FIG. 1 is a system architecture diagram of an intelligent task navigation method based on machine learning according to an embodiment of the present application. Figure 1 and Figure 2 As shown, according to the embodiment of the present application, the intelligent task navigation method based on machine learning includes the steps of: S1, obtaining a task description voice input by a user; S2, performing voice recognition on the task description voice to obtain a task description voice recognition result; S3, performing text semantic understanding analysis on the task description voice recognition result to obtain a set of task description semantic coding feature vectors; S4, passing the set of task description semantic coding feature vectors through a common semantic feature extraction network based on adaptive gating logic to obtain a task description common extraction semantic coding feature vector; S5, determining the intention recognition result based on the task description common extraction semantic coding feature vector; S6, extracting tasks matching the intention recognition result from the task database.

[0033] In particular, the step S1 is to obtain a task description voice input by a user. Here, a more natural interaction relationship can be established with the user through voice input, which reduces the complex operations of the user when performing tasks and provides a smoother and more pleasant use experience.

[0034] In particular, the step S2 performs speech recognition on the task description speech to obtain a task description speech recognition result. In the technical solution of the present application, speech recognition is performed on the task description speech to obtain a task description speech recognition result. Here, the user's speech data can be converted into processable text data through speech recognition, which provides an important data basis for understanding the user's intention.

[0035] In particular, the step S3 performs text semantic understanding analysis on the task description speech recognition result to obtain a set of task description semantic encoding feature vectors. In particular, in a specific example of the present application, Figure 4 As shown, S3 includes: S31, performing data enhancement on the task description speech recognition results to obtain a set of enhanced task description semantic recognition results; S32, performing word segmentation processing on each enhanced task description semantic recognition result in the set of enhanced task description semantic recognition results, and then passing it through a task description semantic encoder including a word embedding layer to obtain a set of task description semantic encoding feature vectors.

[0036] Specifically, the S31 performs data enhancement on the task description speech recognition result to obtain a set of enhanced task description semantic recognition results. It should be understood that although the existing speech recognition method has a high accuracy rate, it may still cause recognition errors due to factors such as the user's pronunciation, speaking speed, accent or background noise. These erroneous texts may have a greater adverse effect on the task of identifying the user's intention. In addition, different words may have the same pronunciation but different meanings (such as the English word "right" can mean "correct" or "right"), and the speech recognition system may not be able to distinguish these homophones, resulting in expression errors. Therefore, in the technical solution of the present application, the task description speech recognition result is data enhanced to alleviate these recognition errors as much as possible and affect subsequent semantic understanding and analysis, thereby obtaining a set of enhanced task description semantic recognition results. In an embodiment of the present application, the data enhancement includes randomly inserting words, deleting or replacing words in the task description speech recognition result. On the one hand, through these data enhancement operations, it can be ensured that the subsequent model can maintain relatively high performance even in the face of noise and imperfect input. On the other hand, these data enhancement operations can enrich the original text data set, thereby increasing the diversity and coverage of user text semantic information. The data enhancement includes randomly inserting words, deleting or replacing words in the task description speech recognition results.

[0037] Specifically, the S32, after word segmentation processing of each enhanced task description semantic recognition result in the set of enhanced task description semantic recognition results, passes through a task description semantic encoder including a word embedding layer to obtain a set of task description semantic encoding feature vectors. Specifically, word segmentation is the process of dividing the text, that is, each enhanced task description semantic recognition result in the set of enhanced task description semantic recognition results, into meaningful units (such as words or phrases). This is an important step in text preprocessing, which enables the model to process elements in the text one by one, rather than treating it as a continuous sequence of characters. Word embeddings can map each enhanced task description semantic recognition word obtained after word segmentation to a high-dimensional space. Usually, a word is converted into a vector representation. These vectors can capture the semantic relationship between words, and similar words are closer in the vector space. Semantic encoding of each vector representation obtained after word embedding through the task description semantic encoder can learn complex semantic information in text data, such as the logical association between each word and the semantic meaning expressed by the text.

[0038] In a specific example of the present application, each enhanced task description semantic recognition result in the set of enhanced task description semantic recognition results is subjected to word segmentation processing and then passed through a task description semantic encoder including a word embedding layer to obtain a set of task description semantic encoding feature vectors, for example:

[0039] Perform word segmentation processing on each enhanced task description semantic recognition result in the set of enhanced task description semantic recognition results to convert each enhanced task description semantic recognition result in the set of enhanced task description semantic recognition results into a word sequence composed of multiple words; use the embedding layer of the task description semantic encoder including the word embedding layer to map each word in the word sequence into a word embedding vector to obtain a sequence of word embedding vectors; use the converter of the task description semantic encoder including the word embedding layer to perform global context semantic encoding on the sequence of word embedding vectors based on the converter idea to obtain a set of task description semantic encoding feature vectors. Specifically, a converter of the task description semantic encoder including a word embedding layer is used to perform global context semantic encoding based on the converter idea on the sequence of word embedding vectors to obtain a set of task description semantic encoding feature vectors, including: arranging the sequence of word embedding vectors in one dimension to obtain a global feature vector; calculating the product between the global feature vector and the transpose vector of each word embedding vector in the sequence of word embedding vectors to obtain a plurality of self-attention association matrices; respectively normalizing each of the plurality of self-attention association matrices to obtain a plurality of standardized self-attention association matrices; passing each of the plurality of standardized self-attention association matrices through a Softmax classification function to obtain a plurality of probability values; respectively weighting the transpose vector of each word embedding vector in the sequence of word embedding vectors using each of the plurality of probability values ​​as a weight to obtain the plurality of context semantic feature vectors; and, cascading the plurality of context semantic feature vectors to obtain the set of task description semantic encoding feature vectors.

[0040] It is worth mentioning that in other specific examples of the present application, the task description speech recognition result can also be subjected to text semantic understanding analysis in other ways to obtain a set of task description semantic encoding feature vectors, for example: inputting the task description speech recognition result; preprocessing the task description speech recognition result; using natural language processing technology to analyze the text, including part-of-speech tagging, named entity recognition, syntactic analysis, etc., to understand the structure and meaning of the text; using semantic analysis technology, such as word embeddings or pre-trained language models (such as BERT, GPT, etc.), to convert the text into a semantic representation; converting the text that has undergone semantic understanding analysis into a set of semantic encoding feature vectors of the task description.

[0041] In particular, the S4 passes the set of the task description semantic coding feature vectors through a common semantic feature extraction network based on adaptive gating logic to obtain a task description common extraction semantic coding feature vector. Considering that the data-enhanced text data will improve the user's semantic expression, it will also add a large amount of redundant information to the system. These redundant information may be retained in the process of semantic encoding, thereby affecting the accurate identification of user intentions. In the technical solution of the present application, the set of the task description semantic coding feature vectors is further passed through a common semantic feature extraction network based on adaptive gating logic to highly abstract and summarize the set of the task description semantic coding feature vectors, extract higher-level semantic information and filter out unimportant or redundant information for understanding user intentions, thereby obtaining a task description common extraction semantic coding feature vector. Specifically, the common semantic feature extraction network based on adaptive gating logic highlights its own contribution and importance by measuring the semantic difference information between each task description semantic encoding feature vector and other task description semantic encoding feature vectors, and combines the gating logic mechanism to screen and distill such contribution and importance, thereby adaptively selecting and emphasizing features that have a greater contribution to the target task, reducing the interference of redundant information and unimportant background information, and suppressing the deviation of feature distribution.

[0042] In a specific example of the present application, the set of task description semantic coding feature vectors is passed through a common semantic feature extraction network based on adaptive gating logic to obtain a task description common extraction semantic coding feature vector, including: calculating the semantic difference determining factors of each task description semantic coding feature vector in the set of task description semantic coding feature vectors relative to other task description semantic coding feature vectors to obtain a sequence of semantic difference determining factors; using a gating logic decider to process the sequence of semantic difference determining factors to obtain a sequence of gated decision semantic difference determining factors; normalizing the sequence of gated decision semantic difference determining factors to obtain a sequence of normalized gated decision semantic difference determining factors; using the sequence of normalized gated decision semantic difference determining factors as weights, calculating the weighted sum of the set of task description semantic coding feature vectors to obtain the task description common extraction semantic coding feature vector. Specifically, calculating the semantic difference determining factors of each task description semantic coding feature vector in the set of task description semantic coding feature vectors relative to other task description semantic coding feature vectors to obtain a sequence of semantic difference determining factors includes: processing each task description semantic coding feature vector in the set of task description semantic coding feature vectors using the following semantic difference determining factor calculation formula to obtain the sequence of semantic difference determining factors; wherein the semantic difference determining factor calculation formula is: ;in, is the first in the sequence of semantic encoding feature vectors of the task description A task description semantic encoding feature vector, is the first in the sequence of semantic encoding feature vectors of the task description A task description semantic encoding feature vector, represents the one-norm of the eigenvector, The value of is the length of the sequence of the semantic encoding feature vector of the task description minus one, a representation of a sequence of semantically encoded feature vectors describing the task, Indicates semantic difference determining factors. The sequence of the semantic difference determining factors is processed using a gated logic decider to obtain a sequence of gated decision semantic difference determining factors, including: setting the characteristic values ​​in the sequence of the semantic difference determining factors that are less than or equal to a predetermined threshold to zero to obtain the sequence of the gated decision semantic difference determining factors. More specifically, the sequence of the gated decision semantic difference determining factors is normalized to obtain a sequence of normalized gated decision semantic difference determining factors, including: processing the sequence of the gated decision semantic difference determining factors using the following normalization formula to obtain the sequence of normalized gated decision semantic difference determining factors; wherein the normalization formula is: ;in, For the The gated decision semantic difference determining factor, is the number of semantic difference determining factors of the gated decision, represents an exponential function with natural numbers as base, For the The normalized gated decision semantic difference determining factor.

[0043] In summary, in the above embodiment, the set of the task description semantic coding feature vectors is passed through a common semantic feature extraction network based on adaptive gating logic to obtain a task description common extraction semantic coding feature vector, including: passing the set of the task description semantic coding feature vectors through a common semantic feature extraction network based on adaptive gating logic to perform common semantic feature extraction according to the following formula to obtain the task description common extraction semantic coding feature vector; wherein the formula is: ;in, is the first in the sequence of semantic encoding feature vectors of the task description A task description semantic encoding feature vector, is the first in the sequence of semantic encoding feature vectors of the task description A task description semantic encoding feature vector, represents the one-norm of the eigenvector, The value of is the length of the sequence of the semantic encoding feature vector of the task description minus one, a representation of a sequence of semantically encoded feature vectors describing the task, Indicates semantic differential determinants, For the The gated decision semantic difference determining factor, is the number of semantic difference determining factors of the gated decision, represents an exponential function with natural numbers as base, For the The normalized gated decision semantic difference determining factor, Indicates masking processing. Extract semantic encoding feature vectors for commonality in the task descriptions.

[0044] In particular, the S5 determines the intent recognition result based on the semantic encoding feature vector extracted from the commonality of the task description. That is, in a specific example of the present application, the semantic encoding feature vector extracted from the commonality of the task description is passed through a classifier-based intent identifier to obtain an intent recognition result, and the intent recognition result is used to represent the task type label. Among them, the classifier can automatically map the user intention features and task semantic descriptions expressed by the semantic encoding feature vector extracted from the commonality of the task description to the task type label with the highest correlation, thereby realizing automated intent recognition. Specifically, the semantic encoding feature vector extracted from the commonality of the task description is passed through a classifier-based intent identifier to obtain an intent recognition result, including: using multiple fully connected layers of the classifier to fully connect the semantic encoding feature vector extracted from the commonality of the task description to obtain an encoded classification feature vector; and, passing the encoded classification feature vector through the Softmax classification function of the classifier to obtain the classification result.

[0045] In particular, the step S6 extracts tasks matching the intention recognition result from the task database. That is, by establishing a task database, after identifying the user's true intention, matching items can be quickly and accurately found from a large number of tasks, which can save the user's time in finding suitable tasks and improve overall work efficiency.

[0046] It should be understood that before using the above-mentioned neural network model for inference, it is necessary to train the task description semantic encoder including the word embedding layer, the common semantic feature extraction network based on adaptive gating logic, and the classifier-based intent identifier. That is to say, in the machine learning-based intelligent task navigation method of the present application, it also includes a training phase for training the task description semantic encoder including the word embedding layer, the common semantic feature extraction network based on adaptive gating logic, and the classifier-based intent identifier.

[0047] Figure 3 FIG. 1 is a flow chart of the training phase of the intelligent task navigation method based on machine learning according to an embodiment of the present application. Figure 3 As shown, according to the embodiment of the present application, the intelligent task navigation method based on machine learning includes: a training phase, including: S110, obtaining training data, wherein the training data includes a training task description voice and a true value of a task type label; S120, performing speech recognition on the training task description voice to obtain a training task description voice recognition result; S130, performing data enhancement on the training task description voice recognition result to obtain a set of training enhanced task description semantic recognition results; S140, performing word segmentation processing on each training enhanced task description semantic recognition result in the set of training enhanced task description semantic recognition results, and then passing it through the task description semantic encoder including the word embedding layer to obtain a set of training task description semantic encoding feature vectors S150, passing the set of semantic encoding feature vectors of the training task description through the common semantic feature extraction network based on adaptive gating logic to obtain the common extraction semantic encoding feature vector of the training task description; S160, passing the common extraction semantic encoding feature vector of the training task description through the classifier-based intent recognizer to obtain a probability score and a classification loss function value; S170, calculating the class probability related loss function value based on the probability score; S180, using the weighted sum of the classification loss function value and the class probability related loss function value as the loss function value to train the task description semantic encoder including the word embedding layer, the common semantic feature extraction network based on adaptive gating logic and the classifier-based intent recognizer.

[0048] In an optimization example, during the model training process, for each iteration, the label probability value corresponding to each task type label is obtained by the classifier of the commonality extraction semantic encoding feature vector of the training task description, and the maximum label probability value is obtained as the probability score, and a class probability related loss function is introduced based on the probability score. The calculation of the class probability related loss function is specifically as follows:

[0049] Calculate one minus the probability score to obtain a score difference;

[0050] Determine the vector mean and vector variance of all feature values ​​of the common extraction semantic encoding feature vector described in the training task as a feature set;

[0051] Calculate the class cross entropy difference obtained by multiplying each eigenvalue of the commonality extraction semantic encoding feature vector of the training task description by the probability score and then subtracting the product of the score difference and the vector mean in the form of class cross entropy;

[0052] Sum the squares of the class cross entropy differences corresponding to each eigenvalue in the commonality extraction semantic encoding feature vector of the training task description and divide the sum by the length of the commonality extraction semantic encoding feature vector of the training task description to obtain a first loss term;

[0053] Multiplying the square sum of the probability score and the score difference by the vector variance to obtain a second loss term;

[0054] The second loss term is subtracted from the first loss term to obtain the class probability related loss function.

[0055] Here, the common semantic feature representation based on adaptive gating logic of the word embedding encoding semantics of each enhanced task description semantic recognition result in the set of enhanced task description semantic recognition results will cause out-of-distribution noise in the multi-dimensional encoding semantic common feature representation set due to the common semantics convergence of the encoding semantics corresponding to different data enhancements in different dimensions, causing class distribution noise in the overall class of the common extracted semantic encoding feature vector of the training task description, thereby affecting the accuracy of the classification result. The above-mentioned class probability related loss function sets the approximate expected variance characteristics of the out-of-distribution noise of the feature set of the common extracted semantic encoding feature vector of the training task description relative to the Bernoulli sampling instance of the feature set based on the class probability distribution characteristics expressed by the probability score and score difference of the common extracted semantic encoding feature vector of the training task description through the classifier.

[0056] Therefore, the class probability related loss function is used to measure the distribution feature noise based on the class probability distribution characteristics expressed by the probability score and score difference of the common extracted semantic coding feature vector described in the training task. When the class probability distribution of each eigenvalue of the common extracted semantic coding feature vector described in the classification process is based on the probabilistic cross random selection characteristics of the expected variance of the common extracted semantic coding feature vector described in the training task, the possible sampling noise characteristics can be degraded according to the probabilistic cross random selection characteristics of the class cross entropy compound difference calculation of the expected variance of the common extracted semantic coding feature vector described in the training task, thereby realizing the class distribution noise reduction of self-supervised learning in the process of updating the model parameters based on the common extracted semantic coding feature vector described in the training task, so as to improve the accuracy of the intent recognition result obtained by the common extracted semantic coding feature vector described in the training task through the classifier-based intent recognizer.

[0057] In summary, according to the embodiment of the present application, an intelligent task navigation method based on machine learning is explained, which establishes an interactive relationship with the user through voice, and then performs voice recognition on the task description voice input by the user to convert it into text data, and performs semantic understanding and analysis on the text data containing user intention information through natural language processing technology based on machine learning, and further extracts the commonalities of semantic features to learn and identify the user's deep semantic meaning, thereby determining the type of task the user wants to perform, and further matching it with the tasks in the task database.

[0058] Furthermore, an intelligent task navigation system based on machine learning is also provided.

[0059] Figure 5 FIG. 1 is a block diagram of an intelligent task navigation system based on machine learning according to an embodiment of the present application. Figure 5 As shown, according to the embodiment of the present application, the intelligent task navigation system 300 based on machine learning includes: an information acquisition module 310, which is used to acquire the task description voice input by the user; a voice recognition module 320, which is used to perform voice recognition on the task description voice to obtain a task description voice recognition result; a text semantic understanding analysis module 330, which is used to perform text semantic understanding analysis on the task description voice recognition result to obtain a set of task description semantic coding feature vectors; a common semantic feature extraction module 340, which is used to pass the set of task description semantic coding feature vectors through a common semantic feature extraction network based on adaptive gating logic to obtain a task description common extraction semantic coding feature vector; a recognition result determination module 350, which is used to determine the intention recognition result based on the task description common extraction semantic coding feature vector; and a task matching module 360, which is used to extract tasks that match the intention recognition result from the task database.

[0060] As described above, the intelligent task navigation system 300 based on machine learning according to the embodiment of the present application can be implemented in various wireless terminals, such as a server with an intelligent task navigation algorithm based on machine learning. In a possible implementation, the intelligent task navigation system 300 based on machine learning according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the intelligent task navigation system 300 based on machine learning can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the intelligent task navigation system 300 based on machine learning can also be one of the many hardware modules of the wireless terminal.

[0061] Alternatively, in another example, the machine learning-based intelligent task navigation system 300 and the wireless terminal may also be separate devices, and the machine learning-based intelligent task navigation system 300 may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0062] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. An intelligent task navigation method based on machine learning, characterized in that: include: Obtaining the task description voice input by the user; Performing speech recognition on the task description speech to obtain a task description speech recognition result; Performing text semantic understanding analysis on the task description speech recognition result to obtain a set of task description semantic encoding feature vectors; Passing the set of task description semantic encoding feature vectors through a common semantic feature extraction network based on adaptive gating logic to obtain a task description common extraction semantic encoding feature vector; Extracting semantic encoding feature vectors based on commonalities of the task descriptions to determine intent recognition results; Extracting tasks matching the intention recognition result from a task database; The set of the task description semantic coding feature vectors is passed through a common semantic feature extraction network based on adaptive gating logic to obtain a task description common extraction semantic coding feature vector, including: Calculating the semantic difference determining factor of each task description semantic encoding feature vector in the set of task description semantic encoding feature vectors relative to other task description semantic encoding feature vectors to obtain a sequence of semantic difference determining factors; Processing the sequence of semantic difference determining factors using a gated logic decider to obtain a sequence of gated decision semantic difference determining factors; Normalizing the sequence of gated decision semantic difference determining factors to obtain a sequence of normalized gated decision semantic difference determining factors; Using the sequence of the normalized gated decision semantic difference determining factors as weights, calculating the weighted sum of the set of task description semantic encoding feature vectors to obtain the task description commonality extraction semantic encoding feature vector; The step of calculating the semantic difference determining factors of each task description semantic coding feature vector in the set of task description semantic coding feature vectors relative to other task description semantic coding feature vectors to obtain a sequence of semantic difference determining factors includes: Each task description semantic coding feature vector in the set of task description semantic coding feature vectors is processed using the following semantic difference determining factor calculation formula to obtain a sequence of the semantic difference determining factors; wherein the semantic difference determining factor calculation formula is: ;in, is the first in the sequence of semantic encoding feature vectors of the task description A task description semantic encoding feature vector, is the first in the sequence of semantic encoding feature vectors of the task description A task description semantic encoding feature vector, represents the one-norm of the eigenvector, The value of is the length of the sequence of the semantic encoding feature vector of the task description minus one, a representation of a sequence of semantically encoded feature vectors describing the task, Indicates semantic differential determinants; The process of using a gated logic decider to process the sequence of semantic difference determining factors to obtain a sequence of gated decision semantic difference determining factors includes: The feature values ​​in the sequence of semantic difference determining factors that are less than or equal to a predetermined threshold are set to zero to obtain the sequence of the gated decision semantic difference determining factors.

2. The intelligent task navigation method based on machine learning according to claim 1, characterized in that: Performing text semantic understanding analysis on the task description speech recognition result to obtain a set of task description semantic encoding feature vectors, including: Performing data enhancement on the task description speech recognition results to obtain a set of enhanced task description semantic recognition results; After word segmentation processing is performed on each enhanced task description semantic recognition result in the set of enhanced task description semantic recognition results, the result is passed through a task description semantic encoder including a word embedding layer to obtain a set of task description semantic encoding feature vectors.

3. The intelligent task navigation method based on machine learning according to claim 2 is characterized in that: The data enhancement includes randomly inserting words, deleting or replacing words in the task description speech recognition result.

4. The intelligent task navigation method based on machine learning according to claim 3 is characterized in that: Normalizing the sequence of the gated decision semantic difference determining factors to obtain a sequence of normalized gated decision semantic difference determining factors, including: The sequence of the gated decision semantic difference determining factors is processed by the following normalization formula to obtain the sequence of the normalized gated decision semantic difference determining factors; wherein the normalization formula is: ;in, For the The gated decision semantic difference determining factor, is the number of semantic difference determining factors of the gated decision, represents an exponential function with natural numbers as base, For the The normalized gated decision semantic difference determining factor.

5. The intelligent task navigation method based on machine learning according to claim 4 is characterized in that: Extracting semantic encoding feature vectors based on the commonality of the task description to determine the intent recognition result includes: The commonality extraction semantic encoding feature vector of the task description is passed through a classifier-based intent identifier to obtain the intent recognition result, and the intent recognition result is used to represent the task type label.

6. The intelligent task navigation method based on machine learning according to claim 5, characterized in that: The method further includes a training step: training the task description semantic encoder including the word embedding layer, the common semantic feature extraction network based on adaptive gating logic, and the classifier-based intent identifier; wherein the training step includes: Acquire training data, wherein the training data includes a training task description voice and a true value of a task type label; Performing speech recognition on the training task description speech to obtain a training task description speech recognition result; Performing data enhancement on the speech recognition results of the training task description to obtain a set of semantic recognition results of the training enhancement task description; After word segmentation processing is performed on each training enhanced task description semantic recognition result in the set of training enhanced task description semantic recognition results, the result is passed through the task description semantic encoder including the word embedding layer to obtain a set of training task description semantic encoding feature vectors; Passing the set of the training task description semantic coding feature vectors through the common semantic feature extraction network based on adaptive gating logic to obtain the training task description common extraction semantic coding feature vector; Passing the commonality extraction semantic encoding feature vector of the training task description through a classifier-based intent recognizer to obtain a probability score and a classification loss function value; Calculating a class probability-related loss function value based on the probability score; The task description semantic encoder including the word embedding layer, the common semantic feature extraction network based on adaptive gating logic and the classifier-based intent identifier are trained using the weighted sum of the classification loss function value and the class probability related loss function value as the loss function value.

7. An intelligent task navigation system based on machine learning, used to execute the intelligent task navigation method based on machine learning according to claim 1, characterized in that: include: An information acquisition module, used to acquire the task description voice input by the user; A speech recognition module, used for performing speech recognition on the task description speech to obtain a task description speech recognition result; A text semantic understanding and analysis module, used for performing text semantic understanding and analysis on the task description speech recognition result to obtain a set of task description semantic encoding feature vectors; A common semantic feature extraction module, used for passing the set of task description semantic encoding feature vectors through a common semantic feature extraction network based on adaptive gating logic to obtain a task description common extraction semantic encoding feature vector; A recognition result determination module, used to extract semantic encoding feature vectors based on the commonality of the task description to determine the intention recognition result; The task matching module is used to extract tasks matching the intention recognition result from the task database.

Images