CN115455151A

CN115455151A - AI emotion visual identification method and system and cloud platform

Info

Publication number: CN115455151A
Application number: CN202211183662.8A
Authority: CN
Inventors: 胡霞; 吴向东
Original assignee: Individual
Current assignee: Individual
Priority date: 2022-09-27
Filing date: 2022-09-27
Publication date: 2022-12-09

Abstract

According to the AI emotion visual identification method, the AI emotion visual identification system and the cloud platform, provided by the embodiment of the invention, an emotional text paragraph and a paragraph word vector corresponding to the emotional text paragraph can be determined according to text big data to be subjected to emotion analysis; and carrying out user emotion analysis by using a target information block and a paragraph word vector in text big data to be subjected to emotion analysis to obtain an emotion analysis report. In this way, the target information block can be used to introduce consideration to the noise emotion so as to improve the accuracy and reliability of emotion analysis.

Description

AI emotion visual identification method and system and cloud platform

Technical Field

The invention relates to the technical field of artificial intelligence, in particular to an AI emotion visual identification method, system and cloud platform.

Background

At present, text mining and analysis of social media, email, chat, product reviews and recommendations has become a valuable resource for almost all industry vertical industry research data models, which can help enterprises obtain more information, learn more about customers, predict and enhance customer experience, customize marketing campaigns, and assist in decision making. Emotion analysis uses a machine learning algorithm to determine the user's emotion to which the textual content corresponds. The emotion analysis use case includes: the conventional emotion analysis technology is easily disturbed by noise emotions, and thus it is difficult to ensure the accuracy of emotion analysis, while the mood of a client review is rapidly known, products or services that the client likes or dislikes are known, factors that may affect a new client's purchase decision are known, a market awareness is provided for an enterprise, and problems are solved as early as possible.

Disclosure of Invention

In order to solve the technical problems in the related art, the invention provides an AI emotion visual identification method, system and cloud platform.

In a first aspect, an embodiment of the present invention provides an AI emotion visualization recognition method, which is applied to an artificial intelligence cloud platform, and the method includes: determining an emotional text paragraph and a paragraph word vector corresponding to the emotional text paragraph according to text big data to be subjected to emotion analysis; and carrying out user emotion analysis by using a target information block and a paragraph word vector in text big data to be subjected to emotion analysis to obtain an emotion analysis report.

In an illustrative embodiment, the determining, according to text big data to be subjected to emotion analysis, an emotional text paragraph and a paragraph word vector corresponding to the emotional text paragraph includes:

obtaining viewpoint comment words of a first user comment information set in text big data to be subjected to emotion analysis; and determining a plurality of first emotional text paragraphs from the first user comment information set by using the obtained viewpoint comment words, and mining a first paragraph word vector of each first emotional text paragraph.

In an exemplary embodiment, the performing emotion analysis on the user by using the target information block and the paragraph word vector in the text big data to be subjected to emotion analysis to obtain an emotion analysis report includes:

determining a marked target information block in the first user comment information set, and determining a marked first emotion analysis noise coefficient of each first emotional text segment in combination with the determined target information block; acquiring a first association score between each first emotional text paragraph determined according to the determined first emotion analysis noise coefficient;

optimizing each first paragraph word vector based on the obtained first association score, and performing joint analysis on the optimized first paragraph word vectors and second paragraph word vectors to obtain an emotion analysis report of the first user comment information set; wherein the second paragraph word vector is: optimizing the paragraph word vector of each second emotional text paragraph according to each second association score to determine a natural language mining vector, wherein each second emotional text paragraph is: and text paragraphs corresponding to the first emotional text paragraphs in a second user comment information set in the reference text big data.

In an exemplary embodiment, the jointly analyzing the optimized first paragraph word vector and the optimized second paragraph word vector to obtain the emotion analysis report of the first user comment information set includes:

determining a commonality index of the optimized first paragraph word vector of each first emotional text paragraph and the corresponding second paragraph word vector as a commonality index corresponding to each first emotional text paragraph;

determining a weighting factor for a first paragraph word vector of each first emotionalized text paragraph for an emotion description vector of the first user comment information set based on a first emotion analysis noise coefficient of each first emotionalized text paragraph;

processing the common indexes corresponding to the first emotional text paragraphs based on the determined weighting factors to obtain processing results, wherein the processing results are used as the common indexes of the first user comment information set and the second user comment information set;

and determining an emotion analysis report of the first user comment information set by using the obtained commonality index.

In one illustrative embodiment, the determining a weighting factor for the first paragraph word vector of each first emotional text paragraph for the emotion description vector of the first user comment information set based on the first emotion analysis noise figure of each first emotional text paragraph comprises: determining a weighting factor for the first paragraph word vector of each first passage of emotional text for the emotion description vector of the first set of user comment information based on the first emotion analysis noise coefficient of each first passage of emotional text and the second emotion analysis noise coefficient of the corresponding second passage of emotional text.

according to the first emotion analysis noise coefficient of each first emotional text paragraph, carrying out vector knowledge splicing on the first paragraph word vectors of each first emotional text paragraph to obtain a first linkage word vector, and according to the second emotion analysis noise coefficient of each second emotional text paragraph, carrying out vector knowledge splicing on the second paragraph word vectors of each second emotional text paragraph to obtain a second linkage verb vector;

determining a commonality index of the first conjunctive word vector and the second conjunctive verb vector; determining an emotional analysis report of the first set of user comment information based on the determined commonality index.

In an exemplary embodiment, the first association score is determined by:

determining the maximum value of the emotion analysis noise coefficient noise _ i and a set variable;

determining the absolute difference value of the emotion analysis noise coefficient noise _ j and 1;

obtaining VALUE _ ij based on the maximum VALUE and the absolute difference VALUE;

the VALUE _ ij is a first association score of the first emotional text paragraph section _ j to the first emotional text paragraph section _ i, the emotion analysis noise coefficient noise _ i represents an emotion analysis noise coefficient of the first emotional text paragraph section _ i, the emotion analysis noise coefficient noise _ j represents an emotion analysis noise coefficient of the first emotional text paragraph section _ j, and the variable is set to be P.

In an exemplary embodiment, the optimizing each first paragraph word vector based on the obtained first association score, and performing joint analysis on the optimized first paragraph word vector and the optimized second paragraph word vector to obtain the emotion analysis report of the first user comment information set includes: loading each first paragraph word vector, a first association score between each first emotional text paragraph, a paragraph word vector of each second emotional text paragraph, and a second association score between each second emotional text paragraph to the natural language processing model which is debugged, so that the natural language processing model optimizes each first paragraph word vector based on each first association score to obtain an optimized first paragraph word vector, optimizes the paragraph word vector of each second emotional text paragraph based on each second association score to obtain a second paragraph word vector, and performs joint analysis on the optimized first paragraph word vector and the optimized second paragraph word vector to generate an emotion analysis report; acquiring the emotion analysis report generated by the natural language processing model;

wherein the determining a number of first emotionalized paragraphs of text from the first set of user comment information using the obtained point of view comment terms comprises: determining a text paragraph label matched with each viewpoint comment word by using the acquired diversified viewpoint labels of the viewpoint comment words and the set mapping characteristics between the viewpoint labels and the text paragraph labels of the first emotional text paragraph; for each text paragraph tag, acquiring a selected distribution variable of an emotional text paragraph matched with the text paragraph tag according to a visual distribution variable of a viewpoint comment word matched with the text paragraph tag, determining a distribution variable difference between the selected distribution variable of the emotional text paragraph matched with the text paragraph tag and a selected distribution variable of an associated emotional text paragraph, and determining a window size of the emotional text paragraph matched with the text paragraph tag based on the determined distribution variable difference; determining a first emotional text paragraph matched with each selected distribution variable based on the obtained selected distribution variable and the determined window size;

wherein the mining of the paragraph word vector for each first emotional text paragraph comprises: mining a text description knowledge relationship network of a first user comment information set in the text big data to be subjected to emotion analysis; determining a local knowledge relationship network corresponding to each first emotional text paragraph in the text description knowledge relationship network based on a distribution variable corresponding to each first emotional text paragraph in the first user comment information set; based on the size of a set relation network, adjusting the local knowledge relation network of each first emotional text paragraph to generate a target knowledge relation network with the size of the set relation network; a paragraph word vector corresponding to the target knowledge relationship network for each first paragraph of emotive text is determined as the paragraph word vector for each first paragraph of emotive text.

In a second aspect, the invention further provides an AI emotion visual recognition system, which includes an artificial intelligence cloud platform and a user terminal, where the artificial intelligence cloud platform is used for: determining an emotional text paragraph and a paragraph word vector corresponding to the emotional text paragraph according to text big data to be subjected to emotion analysis; and carrying out user emotion analysis by using a target information block and a paragraph word vector in text big data to be subjected to emotion analysis to obtain an emotion analysis report.

In a third aspect, the present invention further provides an artificial intelligence cloud platform, including a processor and a memory; the processor is in communication with the memory, and the processor is configured to read the computer program from the memory and execute the computer program to implement the method described above.

In a fourth aspect, the present invention also provides a readable storage medium, on which a program is stored, which when executed by a processor implements the method described above.

Has the beneficial effects that: according to the AI emotion visualization identification method provided by the embodiment of the invention, an emotional text paragraph and a paragraph word vector corresponding to the emotional text paragraph can be determined according to text big data to be subjected to emotion analysis; and carrying out user emotion analysis by using a target information block and a paragraph word vector in text big data to be subjected to emotion analysis to obtain an emotion analysis report. In this way, the target information block can be used to introduce consideration to the noise emotion so as to improve the accuracy and reliability of emotion analysis.

Further, in the AI emotion visualized recognition method provided by the embodiment of the present invention, a plurality of first emotional text paragraphs are determined from the first user comment information set, and then association scores between the first emotional text paragraphs are determined according to emotion analysis noise coefficients of each first emotional text paragraph, where the first association scores of the first emotional text paragraphs are determined based on emotion analysis noise coefficients of each first emotional text paragraph, so that it can be seen that the association scores between the first emotional text paragraphs can reflect noise emotions in the first user comment information set, and when the first paragraph word vectors are optimized by using the first association scores, it can be understood that the first paragraph word vectors of each first emotional text paragraph are optimized based on noise emotions in the first user comment information set, so that interference caused by the noise emotions can be reduced, and thus, an emotion analysis report can be accurately and reliably obtained through joint analysis of the paragraph word vectors.

In addition, the association score between the first emotional text paragraphs is determined through the emotion analysis noise coefficient of each first emotional text paragraph, so that deep mining analysis of mutual influence between the first emotional text paragraphs is realized, emotion recognition analysis is performed on the mutual influence conditions of the emotional text paragraphs obtained through the deep mining analysis, and an emotion analysis report can be accurately and reliably obtained through joint analysis of paragraph word vectors.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description, serve to explain the principles of the invention.

Fig. 1 is a schematic flow chart of an AI emotion visualization recognition method according to an embodiment of the present invention.

Fig. 2 is a schematic communication architecture diagram of an AI emotion visualization recognition system according to an embodiment of the present invention.

Detailed Description

Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary examples do not represent all implementations consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the invention, as detailed in the appended claims.

It should be noted that the terms "first," "second," and the like in the description and claims of the present invention and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order.

The method provided by the embodiment of the invention can be executed in an artificial intelligence cloud platform, computer equipment or a similar arithmetic device. Taking the example of operating on an artificial intelligence cloud platform, the artificial intelligence cloud platform 10 may include one or more processors 102 (the processors 102 may include but are not limited to processing devices such as a microprocessor MCU or a programmable logic device FPGA), and a memory 104 for storing data, and optionally, the artificial intelligence cloud platform may further include a transmission device 106 for communication functions. It will be understood by those of ordinary skill in the art that the above-described structure is merely illustrative and is not intended to limit the structure of the artificial intelligence cloud platform. For example, artificial intelligence cloud platform 10 may also include more or fewer components than shown above, or have a different configuration than shown above.

The memory 104 may be used to store a computer program, for example, a software program and a module of application software, such as a computer program corresponding to an AI emotion visual recognition method in the embodiment of the present invention, and the processor 102 executes the computer program stored in the memory 104, thereby executing various functional applications and data processing, that is, implementing the above method. The memory 104 may include high speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, memory 104 may further include memory located remotely from processor 102, which may be connected to artificial intelligence cloud platform 10 via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The transmission device 106 is used for receiving or transmitting data via a network. Specific examples of the network described above may include a wireless network provided by a communication provider of the artificial intelligence cloud platform 10. In one example, the transmission device 106 includes a Network adapter (NIC), which can be connected to other Network devices through a base station so as to communicate with the internet. In one example, the transmission device 106 may be a Radio Frequency (RF) module, which is used for communicating with the internet in a wireless manner.

Based on this, please refer to fig. 1, fig. 1 is a schematic flow chart of an AI emotion visualization recognition method provided in an embodiment of the present invention, where the method is applied to an artificial intelligence cloud platform, and further includes the following technical solutions described in the following.

The AI emotion visual identification method provided by the embodiment of the invention has the following overall design idea: determining an emotional text paragraph and a paragraph word vector corresponding to the emotional text paragraph according to text big data to be subjected to emotion analysis; and carrying out emotion analysis on the user by using a target information block and a paragraph word vector in the text big data to be subjected to emotion analysis to obtain an emotion analysis report. In this way, the target information block can be used to introduce consideration to the noise emotion so as to improve the accuracy and reliability of emotion analysis.

Under some design ideas which can be independently implemented, determining an emotional text paragraph and a paragraph word vector corresponding to the emotional text paragraph according to text big data to be subjected to emotion analysis; and (4) carrying out emotion analysis on the user by using a target information block and a paragraph word vector in the text big data to be subjected to emotion analysis to obtain an emotion analysis report, wherein the emotion analysis report can be realized based on the steps 101 to 105.

Step 101, obtaining viewpoint comment words of a first user comment information set in text big data to be subjected to emotion analysis.

For example, the text big data to be subjected to emotion analysis may be activity data of a user in various online service interactions or service interaction processes, including text data converted from voice data, or text information formed by communication through characters, expressions, and the like. The text big data may be in the process of e-commerce service, in the process of virtual reality activity, or in the process of hot topic discussion, which is not limited herein.

Further, the user comment information set may include text comment content of the user, and the opinion comment words are key words such as nouns, verbs, adjectives, adverbs, and the like, which constitute the text comment content.

Step 102, determining a plurality of first emotional text paragraphs from the first user comment information set by using the obtained viewpoint comment words, and mining a first paragraph word vector of each first emotional text paragraph.

For example, the emotional text paragraphs may be understood as text paragraphs that may carry emotions of the user, and the emotional text paragraphs may be used for emotion analysis, and generally, the emotional text paragraphs have certain subjective characteristics of the user and may reflect characteristics of the emotional polarity and the like of the user more abundantly. In this regard, a paragraph word vector may be understood as a paragraph text feature of an emotive text paragraph, such as may be obtained by word vector mining through an associated natural language processing model.

And 103, determining marked target information blocks in the first user comment information set, and determining a marked first emotion analysis noise coefficient of each first emotional text segment by combining the determined target information blocks.

For example, a target information block marked in the first user comment information set may be understood as an information set in which various annotations exist in the first user comment information set, and the annotation of the target information block may affect subsequent emotion analysis, so that noise emotion characteristics may be carried in the target information block.

And 104, acquiring a first association score between the first emotional text paragraphs determined according to the determined first emotion analysis noise coefficient.

For example, the first association score may reflect an influence degree between different first emotional text paragraphs, thereby mining the association between different first emotional text paragraphs at a global level to improve the accuracy and credibility of emotion recognition.

And 105, optimizing each first paragraph word vector based on the obtained first association score, and performing joint analysis on the optimized first paragraph word vectors and the optimized second paragraph word vectors to obtain an emotion analysis report of the first user comment information set.

In the embodiment of the present invention, the second paragraph word vector is: optimizing the paragraph word vector of each second emotional text paragraph according to each second association score to determine a natural language mining vector, wherein each second emotional text paragraph is: and text paragraphs corresponding to the first emotional text paragraphs in a second user comment information set in the reference text big data.

For example, performing joint analysis on the optimized first paragraph word vector and the optimized second paragraph word vector may be understood as performing comparative analysis on the optimized first paragraph word vector and the optimized second paragraph word vector to obtain an emotion analysis report of the first user comment information set, where the emotion analysis report may reflect a user emotion polarity corresponding to the first user comment information set, such as a second classification polarity (positive or negative), a third classification polarity (positive, neutral, or negative), or a plurality of classification polarities, and the like, and is not limited herein. In addition, the emotion analysis report can also reflect text contents corresponding to different user emotion polarities, so that the text contents can be subjected to targeted mining analysis to obtain related user requirements or business preferences and the like.

According to the AI emotion visualized identification method provided by the embodiment of the invention, a plurality of first emotional text paragraphs are determined from the first user comment information set, and then association scores among the first emotional text paragraphs are determined according to emotion analysis noise coefficients of the first emotional text paragraphs, because the first association scores of the first emotional text paragraphs are determined based on emotion analysis noise coefficients of the first emotional text paragraphs, based on which the association scores among the first emotional text paragraphs cover noise emotions in the first user comment information set, the process of optimizing the word vectors of the first paragraphs by using the first association scores can be understood as optimizing the word vectors of the first paragraphs based on the noise emotions in the first user comment information set, so that emotion analysis disturbance of the first user comment information set caused by the noise emotions can be reduced, and thus, an emotion analysis report can be accurately and reliably obtained through joint analysis of the word vectors of the paragraphs.

In addition, the common index among the first emotional text paragraphs is determined through the emotion analysis noise coefficient of each first emotional text paragraph, so that the deep mining analysis of the mutual influence among the first emotional text paragraphs is realized, the emotion recognition analysis is performed on the mutual influence of the emotional text paragraphs obtained through the deep mining analysis, and an emotion analysis report can be accurately and reliably obtained through the joint analysis of paragraph word vectors.

For some examples of the present invention, the opinion comment words of the first user comment information set in the text big data to be subjected to emotion analysis may be obtained through an idea of extracting the opinion comment words from the text big data to be subjected to emotion analysis, and an actual extraction idea may be flexibly selected, for example, a debugged opinion comment word extraction network may be used to extract the opinion comment words from the text big data to be subjected to emotion analysis. The opinion comment word extraction network may be various AI models, such as CNN, RNN, DNN, LSTM, and the like. In addition, the viewpoint comment words of the first user comment information set can be manually taken.

For example, the number of viewpoint comment words of the first user comment information set may also be flexibly set, such as the number of viewpoint comment words may be 3: the number of the opinion comment words may be 5: the number of the opinion comment words can be more.

For some examples of the present invention, after viewpoint comment words of an initial user comment information set are extracted, based on the extracted viewpoint comment words, by optimizing wrongly-written characters of words included in the initial user comment information set, an optimized user comment information set is a first user comment information set according to an embodiment of the present invention, and the optimized viewpoint comment words are viewpoint comment words belonging to the first user comment information set in text big data to be subjected to emotion analysis. For example, optimizing the wrongly written words of the word may correct the error based on the context semantics, which is not limited herein.

With respect to step 102, a first passage of emotional text is determined based on the opinion comment terms, which may or may not contain a portion of the opinion comment terms, and may also contain all of the opinion comment terms.

For example, when the opinion comment words of the first user comment information set include: the event opinion comment words, the behavioral opinion comment words, and the content opinion comment words, the determined first emotive text paragraph includes: the event comment information block contains event opinion comment words, the behavior comment information block contains behavior opinion comment words, and the content comment information block contains content opinion comment words. Illustratively, the first emotive text paragraph may also be a first set of user comment information containing event opinion comment words, behavioral opinion comment words, and content opinion comment words.

For some examples of the invention, the first paragraph word vector for mining each first emotional text paragraph may be a feature vector of multiple angles such as punctuation, semantics, and expression of the first emotional text paragraph.

In step 103, the marked target information block in the first user comment information set is the corresponding information block where the noise comment in the first user comment information set is located, such as the information block which is spoken in the normal language and spoken in the reverse language by quotation marks.

The marked target information block in the first user comment information set can be determined through a marked information recognition network (such as a decision tree network) which has completed debugging, and the target information block of the first user comment information set can be manually determined.

Based on the determined target information block in the first user comment information set, a portion, belonging to the target information block, of each first emotional text paragraph can be determined, and then a proportion value of the portion, belonging to the target information block, of each first emotional text paragraph in the first emotional text paragraph can be determined, wherein the determined proportion value can be understood as a first emotion analysis noise coefficient of each first emotional text paragraph marked.

For some examples of the invention, the first emotion analysis noise figure of each first emotional text paragraph may be determined based on words, and after the target information block of the first user comment information set is determined, for example, a first word belonging to the target information block and a second word belonging to the unmarked user comment information set may be determined among words contained in the first user comment information set, so that the number of the first word and the second word contained in each first emotional text paragraph may be determined. For each first passage of emotional text, upon determining a sum of the first number of first words and the second number of second words contained therein, a first emotional analysis noise figure for the first passage of emotional text that is tagged can be determined based on the first number and the second number of the first passage of emotional text.

With respect to step 104, the first emotion analysis noise figure of each first emotional text paragraph may reflect how many noise features are contained in the first paragraph word vector of the first emotional text paragraph, a higher emotion analysis noise figure indicates a higher marked portion of the first emotional text paragraph, and the annotated portion of the first emotional text paragraph negatively contributes to the emotion analysis recognition for the first emotional text paragraph, and thus the more noise features are contained in the first paragraph word vector of the first emotional text paragraph.

It will be appreciated that for any two first passages of emotionalized text, each first passage of emotionalized text reflects a first association score for the other first passage of emotionalized text: the influence of the first paragraph word vector of the first emotional text paragraph on the first paragraph word vector of another first emotional text paragraph is that the higher the emotion analysis noise figure of one first emotional text paragraph is, the lower its first association score to another first emotional text paragraph is, and the higher the emotion analysis noise figure of another first emotional text paragraph is, the lower its first association score to the first emotional text paragraph is.

For some examples of the invention, the first association score between each first emotional text paragraph may be determined based on the following expression:

VALUE _ ij = max (emotion analysis noise coefficient noise _ i, P) × (1-emotion analysis noise coefficient noise _ j)

For step 105, there is a certain degree of association between each first emotional text paragraph, and based on this, the first paragraph word vector of each first emotional text paragraph can be optimized in combination with the degree of association between each first emotional text paragraph.

Illustratively, for an emotive text paragraph, the strictness of the optimization of the first paragraph word vector of the first emotive text paragraph is proportional to the first association score between the first emotive text paragraph and other first emotive text paragraphs, and the strictness of the optimization is higher the first association score is.

For each group of text big data to be subjected to emotion analysis of emotion analysis, in order to analyze user emotion information corresponding to a user comment information set, a group of priori reference text big data of the user emotion information needs to be used for joint analysis with the reference text big data, whether user emotions corresponding to a first user comment information set in the text big data to be subjected to emotion analysis and a second user comment information set in the reference text big data are consistent or not is judged, and if the user emotions are consistent, the priori user emotion information is used as user emotion information of the user emotion corresponding to the text big data to be subjected to emotion analysis, and emotion analysis is completed.

For the second user comment information set in the reference text big data, the second emotional text paragraph, the paragraph word vector of the second emotional text paragraph, the second association score between the second emotional text paragraphs, and the second paragraph word vector obtained by optimizing the paragraph word vector based on the second association score may be determined in advance.

Illustratively, the reference text big data is processed based on steps 101 to 104, and a second paragraph word vector of a second emotional text paragraph of the second user comment information set in the reference text big data is obtained.

In some examples, for each first passage of emotive text, the second passage of emotive text corresponding to the first passage of emotive text may be: a second emotional text paragraph whose distribution variable corresponding to the second user comment information set is consistent with the distribution variable corresponding to the first emotional text paragraph in the first user comment information set, or a second emotional text paragraph whose text paragraph label is consistent with the text paragraph label (such as paragraph type) of the first emotional text paragraph, or a second emotional text paragraph meeting the above two requirements at the same time.

For example, when the paragraph label of a first emotional paragraph of text is a public opinion paragraph label, the second emotional paragraph of text corresponding to the first emotional paragraph of text may be: the text paragraph label within the second user comment information set is a second emotionalized text paragraph of the public opinion perspective paragraph label.

For some examples of the invention, step 105 may be implemented based on a natural language processing model that has been debugged, and may illustratively include the following: loading each first paragraph word vector, a first association score between each first emotional text paragraph, a paragraph word vector of each second emotional text paragraph, and a second association score between each second emotional text paragraph to the natural language processing model which is debugged, so that the natural language processing model optimizes each first paragraph word vector based on each first association score to obtain a first paragraph word vector which is optimized, optimizes the paragraph word vector of each second emotional text paragraph based on each second association score to obtain a second paragraph word vector, and performs joint analysis on the first paragraph word vector and the second paragraph word vector which are optimized to generate an emotion analysis report; an emotion analysis report generated by a natural language processing model (NLP model) is obtained.

Illustratively, a first paragraph word vector of each first emotional text paragraph, a paragraph word vector of each second emotional text paragraph, and a variable distribution list (such as a variable matrix) in which a first association score between the first emotional text paragraphs and a second association score between the second emotional text paragraphs are recorded are used as inputs to the natural language processing model. In some examples, to obtain more accurate and complete information, the variable distribution list may further include an association score between the first passages of emotional text and the second passages of emotional text.

For some examples of the present invention, the natural language processing model may adopt different architectures, and exemplarily, a convolutional neural network may be used to perform deployment of the natural language processing model, and the configuration of the natural language processing model may also be implemented based on a feature pyramid network

In combination with the related contents, the embodiment of the present invention further provides an emotion analysis method, which can be implemented independently, wherein the performing step 105 includes the following steps.

Step 201, determining a commonality index of the optimized first paragraph word vector of each first emotional text paragraph and the corresponding second paragraph word vector as a commonality index corresponding to each first emotional text paragraph.

The common index of the optimized first paragraph word vector and the optimized second paragraph word vector may be determined according to a vector distribution relationship network (e.g., a feature matrix) corresponding to the first paragraph word vector and the second paragraph word vector, and the common index of the optimized first paragraph word vector and the optimized second paragraph word vector is determined by determining a cosine similarity value of the vector distribution relationship network corresponding to the first paragraph word vector and the vector distribution relationship network corresponding to the second paragraph word vector, where, for example, the higher the cosine similarity value is, the higher the common index is.

Step 202, based on the first emotion analysis noise coefficient of each first emotional text paragraph, determining a weighting factor of the first paragraph word vector of each first emotional text paragraph for the emotion description vector of the first user comment information set.

It will be appreciated that the higher the first emotion analysis noise figure, the more noise features are contained in the first paragraph word vector of the first emotional text paragraph, and that the higher the first emotion analysis noise figure, the lower the weighting factor (weight) of the first paragraph word vector of the first emotional text paragraph against the emotion description vector of the first user comment information set, as a whole.

For example, a mapping characteristic (such as a mapping or correspondence between the emotion analysis noise coefficient and the weighting factor) of the emotion analysis noise coefficient and the weighting factor may be determined in advance, and after the first emotion analysis noise coefficient of each first emotional text paragraph is determined, the weighting factor to which the first emotion analysis noise coefficient of each first emotional text paragraph is corresponding may be determined based on the mapping.

For example, each passage of emotionalized text may be assigned a value based on a first emotion analysis noise figure of each passage of emotionalized text, and a ratio of variable values of each passage of emotionalized text may be used as a weighting factor for each first passage of emotionalized text.

For example, a weighting factor of the first paragraph word vector of each first emotional text paragraph for the emotion description vector of the first user comment information set may also be determined based on the first emotion analysis noise coefficient of each first emotional text paragraph and the second emotion analysis noise coefficient of the corresponding second emotional text paragraph.

And 203, processing the commonality indexes corresponding to the first emotional text paragraphs based on the determined weighting factors to obtain a processing result, wherein the processing result is used as the commonality index of the first user comment information set and the second user comment information set.

For example, the first user comment information set includes three first emotional text paragraphs, which are T1, T2, and T3, the commonality index of T1 is 0.8, the weighting factor is 0.5, the commonality index of T2 is 0.5, the weighting factor is 0.2, the commonality index of T3 is 0.9, the weighting factor is 0.3, and the commonality index of the first user comment information set and the second user comment information set is: 0.5+ 0.8+0.2 + 0.5+0.3 + 0.9=0.77.

And step 204, determining an emotion analysis report of the first user comment information set by using the obtained commonality index.

For example, a commonality index determination value may be set, and when the obtained commonality index is greater than the commonality index determination value, it is determined that the emotion vector of the first user comment information set is consistent with the emotion vector of the second user comment information set.

Based on the related content, the idea of performing joint analysis on the optimized first paragraph word vectors and the optimized second paragraph word vectors is provided, and the common index of each first paragraph word vector is independently determined, so that the operation cost of each determination can be reduced.

Under some independently implementable design considerations, performing step 105 may include the following.

Step 301, according to the first emotion analysis noise coefficient of each first emotional text paragraph, performing vector knowledge stitching on the first paragraph word vectors of each first emotional text paragraph to obtain a first linkage word vector, and according to the second emotion analysis noise coefficient of each second emotional text paragraph, performing vector knowledge stitching on the second paragraph word vectors of each second emotional text paragraph to obtain a second linkage verb vector.

The vector knowledge concatenation of the word vectors of the first paragraphs of each first emotional text paragraph may be performed based on the first emotion analysis noise coefficients of the first emotional text paragraphs, for example, the higher the first emotion analysis noise coefficient is, the lower the proportional value example of the first emotional text paragraph is when performing feature fusion, and the lower the first emotion analysis noise coefficient is, the higher the proportional value example of the first emotional text paragraph is when performing feature fusion. It will be appreciated that vector knowledge stitching of the second paragraph word vectors of each second emotionalized text paragraph is similar to the stitching of the first paragraph word vectors of each first emotionalized text paragraph.

And step 302, determining a commonality index of the first linkage word vector and the second linkage verb vector.

The common index of the first linkage word vector and the second linkage verb vector can be determined through the difference of a vector distribution relation network corresponding to the first linkage word vector and a vector distribution relation network corresponding to the second linkage verb vector on a numerical level, and when the difference of the vector distribution relation network corresponding to the first linkage word vector and the vector distribution relation network corresponding to the second linkage verb vector on the numerical level is lower, the common index of the first linkage word vector and the second linkage verb vector is higher.

Step 303, determining an emotion analysis report of the first user comment information set based on the determined commonality index.

By the design, the characteristic fusion is firstly carried out to obtain the linkage word vector, and then the commonality index determination is carried out on the linkage verb vector (the fused characteristic), so that the processing steps can be reduced, and the calculation timeliness is improved.

For some examples of the present invention, the emotion analysis method described above, based on joint analysis of the optimized first paragraph word vector and the optimized second paragraph word vector, may be implemented based on a set AI model to obtain an emotion analysis report of the first user comment information set.

In combination with the emotion analysis method, the following may be included to determine the first emotional text paragraph.

Step 401, determining a text paragraph label matched with each viewpoint comment word by using the obtained diversified viewpoint labels of the viewpoint comment words and the set mapping characteristics between the viewpoint labels and the text paragraph labels of the first emotional text paragraph.

The mapping feature between the set viewpoint label and the text paragraph label of the first emotional text paragraph may be a one-to-one matching relationship, and the exemplary set mapping feature may also be a plurality of viewpoint labels corresponding to one text paragraph label, or a plurality of viewpoint labels corresponding to a plurality of text paragraph labels. Exemplary mapping features may be flexibly set.

Step 402, for each paragraph label, obtaining a selected distribution variable of the emotional paragraph of text matched to the paragraph label according to the visual distribution variable of the viewpoint comment word matched to the paragraph label, determining a distribution variable difference between the selected distribution variable of the emotional paragraph of text matched to the paragraph label and the selected distribution variable of the associated emotional paragraph of text, and determining a window size of the emotional paragraph of text matched to the paragraph label based on the determined distribution variable difference.

For each text paragraph label, a global distribution vector of the viewpoint comment words matched with the text paragraph label can be determined, and a feature value corresponding to the determined global distribution vector is used as a selected distribution variable of the emotional text paragraph matched with the text paragraph label. After determining the selected distribution variable of the emotional text paragraphs of each text paragraph tag, the distribution variable difference of the associated emotional text paragraphs may be determined, wherein the higher the distribution variable difference is, the higher the window size is, and the window size may be the line length and the column length of the emotional text paragraphs.

Step 403, determining a first emotional text paragraph matched with each selected distribution variable based on the obtained selected distribution variables and the determined window size.

The first emotional text paragraph may be a rectangular paragraph, and at this time, a distribution variable of the first emotional text paragraph may be determined based on the determined selected distribution variable and the window size.

Based on the content, the first emotional text paragraph can be determined by combining the content coverage range of each user comment information set, so that the determined first emotional text paragraph is more reasonable, and the emotion analysis precision is further improved.

Under some design ideas which can be independently implemented, the mining of the first paragraph word falling vector can be realized and comprises the following contents.

Step 501, a text description knowledge relationship network of a first user comment information set in text big data to be subjected to emotion analysis is mined.

The text description knowledge relationship network of the first user comment information set may be a semantic feature set, a punctuation feature set, and the like of the first user comment information set, and is not limited herein.

Illustratively, mining of a text description knowledge relationship network of a first user comment information set in text big data to be subjected to emotion analysis can be realized based on a description vector mining submodel of a deep learning model, and the mined text description knowledge relationship network is the feature distribution (target knowledge relationship network) of the first user comment information set.

Step 502, determining a local knowledge relationship network corresponding to each first emotional text segment in the text description knowledge relationship network based on a distribution variable corresponding to each first emotional text segment in the first user comment information set.

Each word in the first user comment information set is located at a position corresponding to the text description knowledge relationship network in the text description knowledge relationship network, so that the local knowledge relationship network of each first emotional text paragraph projected on the text description knowledge relationship network can be determined according to the matching condition between the first user comment information set and the text description knowledge relationship network.

Step 503, based on the size of the set relationship network, adjusting the local knowledge relationship network of each first emotional text paragraph to generate a target knowledge relationship network with the size of the set relationship network.

Wherein, the local knowledge relationship network with different sizes of each first emotional text paragraph can be adjusted to a target knowledge relationship network with a constant size based on the feature pooling processing.

Step 504, determining a paragraph word vector corresponding to the target knowledge relationship network of each first emotional text paragraph as a paragraph word vector in each first emotional text paragraph.

The target knowledge relationship network obtained in step 503 may be processed further to obtain paragraph word vectors in each first emotional text paragraph.

For some examples of the invention, after deep learning model processing, a target knowledge relationship network of user comment information sets in text big data to be subjected to emotion analysis is obtained, pooling processing is performed to obtain a plurality of target knowledge relationship networks with consistent sizes, and one-dimensional paragraph word vectors are obtained through feature integration processing.

Therefore, the first paragraph word vectors of the first emotional text paragraphs can be flexibly determined, and the operation overhead is effectively reduced.

Under some design ideas which can be independently implemented, after optimizing each first paragraph word vector based on the obtained first association score and performing joint analysis on the optimized first paragraph word vector and the optimized second paragraph word vector to obtain an emotion analysis report of the first user comment information set, the method may further include the following steps: determining a user comment paragraph corresponding to the first user comment information set according to the positive emotion polarity information in the emotion analysis report; mining to obtain user requirements based on the user comment paragraphs; and pushing the big data according to the user requirement.

For example, positive emotion polarity information is firstly located, and then the positive emotion polarity information is matched with a user comment paragraph in a first user comment information set, so that accurate user comment paragraph screening can be achieved, user demand analysis of the user comment paragraph is performed in a targeted manner, and personalized big data push (such as topic push, product push and the like) is performed according to the obtained user demand.

Under some design ideas which can be independently implemented, the user requirements are mined based on the user comment paragraphs, and the user requirements can include the following contents: loading the user comment paragraph to an intention feature extraction unit (such as a feature extraction module) in a demand decision analysis model (such as a decision tree model trained in advance), and obtaining a first intention description field and a second intention description field of the user comment paragraph output by the intention feature extraction unit, wherein the intention feature extraction unit includes a plurality of sliding filter subunits (such as a convolution module) connected in sequence, the first intention description field is an intention description field output by a non-end sliding filter subunit of the plurality of sliding filter subunits connected in sequence, and the second intention description field is an intention description field output by an end sliding filter subunit of the plurality of sliding filter subunits connected in sequence; loading the second intention description field to a preference prediction unit in the demand decision analysis model to obtain a preference prediction result output by the preference prediction unit; and loading the first intention description field, the second intention description field, a third intention description field and the preference prediction result to a requirement matching unit in the requirement decision analysis model to obtain the user requirement characteristics output by the requirement matching unit, wherein the third intention description field is an intention description field output by a sliding filter subunit in the preference prediction unit according to an auxiliary description field, and the auxiliary description field is a description field obtained by updating the second intention description field.

In the embodiment of the invention, the layer-by-layer progressive requirement mining processing can be carried out based on the preference prediction unit and the requirement matching unit, so that the requirement mining matching can be carried out from coarse to fine, and the integrity and the accuracy of the requirement characteristics of the user are ensured.

Based on the same or similar inventive concepts, please refer to fig. 2 in combination, and an architecture schematic diagram of an AI emotion visualization recognition system 30 is further provided, which includes an artificial intelligence cloud platform 10 and a user terminal 20 that are in communication with each other, and the artificial intelligence cloud platform 10 and the user terminal 20 implement or partially implement the technical solutions described in the above method embodiments when running.

Further, a readable storage medium is provided, on which a program is stored which, when being executed by a processor, carries out the above-mentioned method.

In the embodiments provided in the embodiments of the present invention, it should be understood that the disclosed apparatus and method may be implemented in other manners. The apparatus and method embodiments described above are merely illustrative and, for example, the flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

In addition, the functional modules in the embodiments of the present invention may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

The functions may be stored in a computer-readable storage medium if they are implemented in the form of software functional modules and sold or used as separate products. Based on such understanding, the technical solution of the present invention or a part thereof which substantially contributes to the prior art may be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a media service server 10, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk, and various media capable of storing program codes. It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising a … …" does not exclude the presence of another identical element in a process, method, article, or apparatus that comprises the element.

The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An AI emotion visual identification method is applied to an artificial intelligence cloud platform, and the method comprises the following steps:

determining an emotional text paragraph and a paragraph word vector corresponding to the emotional text paragraph according to the text big data to be subjected to emotion analysis;

and carrying out user emotion analysis by using a target information block and a paragraph word vector in text big data to be subjected to emotion analysis to obtain an emotion analysis report.

2. The method of claim 1, wherein determining an emotional text paragraph and a paragraph word vector corresponding to the emotional text paragraph according to text big data to be subjected to emotion analysis comprises:

3. The method of claim 2, wherein the performing emotion analysis on the user by using the target information block and the paragraph word vector in the text big data to be subjected to emotion analysis to obtain an emotion analysis report comprises:

optimizing each first paragraph word vector based on the obtained first association score, and performing joint analysis on the optimized first paragraph word vectors and second paragraph word vectors to obtain an emotion analysis report of the first user comment information set; wherein the second paragraph word vector is: optimizing the paragraph word vector of each second emotional text paragraph according to each second association score to determine a natural language mining vector, wherein each second emotional text paragraph is: and text paragraphs corresponding to the first emotional text paragraphs in the second user comment information set in the reference text big data.

4. The method of claim 2, wherein jointly analyzing the optimized first paragraph word vector and the optimized second paragraph word vector to obtain an emotion analysis report of the first user comment information set comprises:

5. The method of claim 1, wherein determining a weighting factor for a first paragraph word vector of each first emotionally-emotioned text paragraph for an emotive description vector of the first user comment information set based on the first emotionally-analyzed noise figure for each first emotionally-typed text paragraph comprises: determining a weighting factor for the first paragraph word vector of each first passage of emotional text for the emotion description vector of the first set of user comment information based on the first emotion analysis noise coefficient of each first passage of emotional text and the second emotion analysis noise coefficient of the corresponding second passage of emotional text.

6. The method of claim 3, wherein jointly analyzing the optimized first paragraph word vector and the optimized second paragraph word vector to obtain an emotion analysis report of the first user comment information set comprises:

determining a commonality index of the first conjunctive word vector and the second conjunctive verb vector; determining an emotional analysis report for the first set of user review information based on the determined commonality index.

7. The method of claim 3, wherein the first association score is determined by:

the VALUE _ ij is a first association score of a first emotional text paragraph section _ j to the first emotional text paragraph section _ i, the emotion analysis noise coefficient noise _ i represents an emotion analysis noise coefficient of the first emotional text paragraph section _ i, the emotion analysis noise coefficient noise _ j represents an emotion analysis noise coefficient of the first emotional text paragraph section _ j, and the variable is set to be P.

8. The method of claim 1, wherein optimizing each first paragraph word vector based on the obtained first association score and performing joint analysis on the optimized first paragraph word vector and the optimized second paragraph word vector to obtain an emotion analysis report of the first user comment information set comprises: loading each first paragraph word vector, a first association score between each first emotional text paragraph, a paragraph word vector of each second emotional text paragraph, and a second association score between each second emotional text paragraph to the natural language processing model which is debugged, so that the natural language processing model optimizes each first paragraph word vector based on each first association score to obtain an optimized first paragraph word vector, optimizes the paragraph word vector of each second emotional text paragraph based on each second association score to obtain a second paragraph word vector, and performs joint analysis on the optimized first paragraph word vector and the optimized second paragraph word vector to generate a sentiment analysis report; acquiring the emotion analysis report generated by the natural language processing model;

9. The AI emotion visual identification system is characterized by comprising an artificial intelligence cloud platform and a user terminal which are communicated with each other, wherein the artificial intelligence cloud platform is used for: determining an emotional text paragraph and a paragraph word vector corresponding to the emotional text paragraph according to text big data to be subjected to emotion analysis; and carrying out user emotion analysis by using a target information block and a paragraph word vector in text big data to be subjected to emotion analysis to obtain an emotion analysis report.

10. An artificial intelligence cloud platform comprising a processor and a memory; the processor is connected in communication with the memory, and the processor is configured to read the computer program from the memory and execute the computer program to implement the method of any one of claims 1 to 8.

11. A readable storage medium, characterized in that the package has stored thereon a program which, when executed by a processor, carries out the method of any one of the preceding claims 1 to 8.