CN110929528A - Statement emotion analysis method, device, server and storage medium - Google Patents
Statement emotion analysis method, device, server and storage medium Download PDFInfo
- Publication number
- CN110929528A CN110929528A CN201911147402.3A CN201911147402A CN110929528A CN 110929528 A CN110929528 A CN 110929528A CN 201911147402 A CN201911147402 A CN 201911147402A CN 110929528 A CN110929528 A CN 110929528A
- Authority
- CN
- China
- Prior art keywords
- vocabularies
- vocabulary
- word
- information vector
- vector
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
The application provides a method, a device, a server and a storage medium for analyzing sentence emotion, and belongs to the technical field of natural language processing. The method comprises the following steps: determining a first word vector corresponding to a plurality of words included in a sentence to be analyzed; processing first word vectors corresponding to the vocabularies, dependency relationships among the vocabularies and types of the dependency relationships on the vocabularies on the basis of a graph convolution network to obtain second word vectors corresponding to the vocabularies; and determining the emotion type of at least one first target vocabulary in the plurality of vocabularies based on the second word vectors corresponding to the plurality of vocabularies. Because the dependency relationship among the vocabularies and the type of each dependency relationship are also considered when the second word vectors corresponding to the vocabularies are determined based on the graph convolution network, even if the sentence to be analyzed is long, the relation among the vocabularies can be strengthened through the dependency relationship among the vocabularies, and the precision of the analysis result of the emotion type of each entity vocabulary is improved.
Description
Technical Field
The present application relates to the field of natural language processing technologies, and in particular, to a method, an apparatus, a server, and a storage medium for analyzing sentence emotion.
Background
Emotion analysis is a common scenario in natural language processing, and can be used for guiding iterative update of products, improving the level of service, and the like. For example, for meal evaluation in daily life, entity recognition is carried out on evaluation contents, and emotion types corresponding to the entities are analyzed, so that emotion information of a user for multiple dimensions such as dish taste, serving time, service attitude and meal environment can be obtained, and the quality degree of each aspect (aspect-level) at the present stage can be determined based on the emotion information, and therefore, how to improve the quality degree is determined.
In the related art, there is an emotion analysis method based on deep learning. The specific implementation steps are as follows: firstly, carrying out word vector coding on each vocabulary contained in a sentence to be analyzed, then carrying out feature extraction based on a convolutional neural network, and finally outputting the probability that each vocabulary of the words belongs to the entity vocabulary, thereby obtaining the probability of the emotion category to which each entity vocabulary belongs.
The related art has a problem that when a sentence to be analyzed is long, a distance between words having an association relationship in the sentence is large, and thus accuracy of an analysis result is affected.
Disclosure of Invention
The embodiment of the application provides a method, a device, a server and a storage medium for analyzing sentence emotion, which are used for solving the problem that when a sentence to be analyzed is long, the distance between words with an association relation in the sentence is large, and therefore the accuracy of an analysis result is influenced. The technical scheme is as follows:
in one aspect, a method for analyzing emotion of a sentence is provided, including:
determining a first word vector corresponding to a plurality of words included in a sentence to be analyzed;
processing first word vectors corresponding to the vocabularies, dependency relations among the vocabularies and types of the dependency relations on the vocabularies on the basis of a graph convolution network to obtain second word vectors corresponding to the vocabularies;
and determining the emotion type to which at least one first target vocabulary in the vocabularies belongs based on the second word vectors corresponding to the vocabularies.
In another aspect, an apparatus for analyzing emotion of a sentence is provided, including:
the system comprises a determining module, a judging module and a judging module, wherein the determining module is used for determining first word vectors corresponding to a plurality of vocabularies included in a sentence to be analyzed;
the processing module is used for processing the first word vectors corresponding to the vocabularies, the dependency relationships among the vocabularies and the types of the dependency relationships on the vocabularies on the basis of a graph convolution network to obtain second word vectors corresponding to the vocabularies;
the determining module is further configured to determine an emotion type to which at least one first target vocabulary in the multiple vocabularies belongs based on the second word vectors corresponding to the multiple vocabularies.
In a possible implementation manner, the determining module is further configured to perform word segmentation on the sentence to be analyzed to obtain a plurality of words; for any vocabulary, splicing word vectors corresponding to the first meanings of the vocabulary and word vectors corresponding to the second meanings of the vocabulary to obtain the first word vectors corresponding to the vocabulary.
In another possible implementation manner, the processing module is further configured to obtain a dependency relationship between the plurality of vocabularies; determining a relation matrix according to the dependency relation, wherein elements in the relation matrix are used for indicating whether dependency relation exists among all vocabularies; for any dependency relationship, acquiring a relationship vector corresponding to the dependency relationship according to the type of the dependency relationship; and processing the relation matrix and the plurality of relation vectors based on the graph convolution network to obtain second word vectors corresponding to the plurality of words.
In another possible implementation manner, the processing module is further configured to, for any vocabulary in the plurality of vocabularies, obtain a first word vector corresponding to at least one vocabulary having a dependency relationship with the vocabulary; and inputting the relation matrix, the plurality of relation vectors and the first word vector corresponding to the at least one vocabulary into the graph convolution network to obtain a second word vector corresponding to the vocabulary.
In another possible implementation manner, the determining module is further configured to process the first word vectors corresponding to the multiple vocabularies based on a first convolutional neural network to obtain third word vectors corresponding to the multiple vocabularies, where the third word vectors correspond to the second word vectors in a one-to-one manner; splicing the corresponding second word vector and the third word vector to obtain a fourth word vector corresponding to the plurality of words; and determining the emotion type to which at least one first target vocabulary in the vocabularies belongs based on the fourth word vectors corresponding to the vocabularies.
In another possible implementation manner, the determining module is further configured to determine, according to a second word vector corresponding to the multiple words, a first information vector corresponding to the multiple words and a second information vector corresponding to the multiple words, where the first information vector is used to indicate whether any of the multiple words is a first target word, and the second information vector is used to indicate an emotion type to which each of the multiple words belongs; and for each vocabulary, when the probability that the vocabulary belongs to a first target vocabulary is determined to be greater than a target probability threshold value according to a first information vector corresponding to the vocabulary, determining the probability that the vocabulary belongs to each emotion type according to a second information vector corresponding to the vocabulary.
In another possible implementation manner, the determining module is further configured to re-encode second word vectors corresponding to the multiple vocabularies based on a second convolutional neural network to obtain first information vectors corresponding to the multiple vocabularies; recoding second word vectors corresponding to the vocabularies on the basis of a third convolutional neural network to obtain second information vectors corresponding to the vocabularies; splicing the first information vector and the second information vector to obtain a third information vector; and respectively recoding the third information vector based on the second convolutional neural network and the third convolutional neural network to obtain the updated first information vector and second information vector.
In another possible implementation manner, the apparatus further includes:
the splicing module is used for splicing the updated first information vector and the updated second information vector to obtain an updated third information vector;
the encoding module is used for respectively re-encoding the updated third information vector again based on the second convolutional neural network and the third convolutional neural network;
and the updating module is used for repeating the splicing and recoding processes until the updating stopping condition is reached.
In another possible implementation manner, the determining module is further configured to obtain interaction transfer information obtained by the second convolutional neural network according to a second word vector corresponding to the multiple words, where the interaction transfer information is used to indicate a word belonging to a second target word in the multiple words; recoding the third information vector based on the third convolutional neural network and the interactive transfer information to obtain an updated second information vector; and recoding the third information vector based on the second convolutional neural network to obtain an updated first information vector.
In another aspect, a server is provided, which includes a processor and a memory, where the memory is used to store at least one program code, and the at least one program code is loaded and executed by the processor to implement the operations executed in the method for analyzing emotion of sentences in the embodiment of the present application.
In another aspect, a storage medium is provided, and the storage medium stores at least one program code, and the at least one program code is used for being executed by a processor and implementing the method for analyzing sentence emotion in the embodiment of the present application.
The technical scheme provided by the embodiment of the application has the following beneficial effects:
in the embodiment of the application, when the second word vectors corresponding to the vocabularies are determined for the first word vectors corresponding to the vocabularies based on the graph convolution network, the dependency relationships among the vocabularies and the types to which the dependency relationships belong are also considered, so that even if the sentence to be analyzed is long, the relation among the vocabularies can be strengthened through the dependency relationships among the vocabularies, and the accuracy of the analysis result of the emotion types to which the entity vocabularies belong is improved.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
FIG. 1 is a block diagram of a sentence emotion analysis system provided in an embodiment of the present application;
FIG. 2 is a flowchart of a method for emotion analysis of a sentence according to an embodiment of the present application;
FIG. 3 is a diagram illustrating a vocabulary dependency according to an embodiment of the present application;
FIG. 4 is a schematic diagram of an architecture of a method for emotion analysis of sentences according to an embodiment of the present application;
FIG. 5 is a block diagram of an apparatus for emotion analyzing of sentences according to an embodiment of the present application;
fig. 6 is a schematic structural diagram of a server according to an embodiment of the present application.
Detailed Description
To make the objects, technical solutions and advantages of the present application more clear, embodiments of the present application will be described in further detail below with reference to the accompanying drawings.
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 embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present application, as detailed in the appended claims.
The embodiment of the application relates to a scene needing emotion analysis on sentences. Correspondingly, the method can be applied to scenes in daily life, for example, a customer can issue a meal evaluation after eating, a customer can issue a meal delivery evaluation after taking out, and a customer can issue a commodity evaluation after purchasing commodities. The evaluations have positive evaluations, such as enjoyable eating, good environment, fast food delivery or good commodity use, negative evaluations, such as difficult eating, high price, slow serving, cool takeaway or rough commodity doing, and neutral evaluations, such as returning, returning and the like. By performing emotion analysis on the sentences in the evaluation, the restaurant can be helped to improve dishes and services, or a takeout service provider can be helped to improve meal delivery services, or commodities can be helped to be upgraded and updated. Correspondingly, the method can also be applied to scenes in the Internet, for example, a user can comment on a movie or a television play, and the user's preference degree on the movie or the television play can be known through analyzing the comment content, so that the playing time is specified; the user can also issue opinions on the hot events, and through the analysis of the opinion content, related departments can be helped to know the opinions of citizens on the hot events; the user can also share the mood of the user, and the mood of the user can be known through the analysis of the mood content. Accordingly, the method can also be applied to other scenes, and is not listed here. Therefore, the embodiment of the application provides a method for analyzing the emotion of the sentences.
The main flow of the method for analyzing emotion of sentences provided by the embodiment of the present application is briefly described below. First word vectors corresponding to a plurality of vocabularies included in a sentence to be analyzed are determined, and then the first word vectors, the dependency relationships among the vocabularies and the types of the dependency relationships are processed based on a graph convolution network, so that second word vectors corresponding to the vocabularies are obtained. And finally, based on the second word vectors corresponding to the vocabularies, the emotion type of at least one first target vocabulary in the vocabularies can be determined, and the emotion analysis of all aspects of the sentence to be analyzed is realized.
Fig. 1 is a structural block diagram of a statement emotion analysis system 100 provided in an embodiment of the present application, where the statement emotion analysis system 100 may be used to implement emotion analysis on a statement, and includes: a plurality of terminals 110 and a sentence emotion analysis platform 120.
The terminal 110 may be connected to the sentence emotion analysis platform 120 through a wireless network or a wired network. The terminal 110 may be at least one of a smartphone, a camcorder, a desktop computer, a tablet computer, an MP4 player, and a laptop portable computer. The terminal 110 is installed and operated with an application program including a sentence uploading function, such as a takeout program, a shopping program, a comment program, and the like. Illustratively, the terminal 110 may be a terminal used by a user, and an account of the user is logged in an application program run by the terminal.
The sentence emotion analysis platform 120 includes at least one of a server, a plurality of servers, and a cloud computing platform. The sentence emotion analysis platform 120 is used for providing background services of the application program including the sentence uploading function, such as sentence segmentation, vocabulary extraction, sentence emotion analysis, and the like. In one possible approach, the sentence emotion analysis platform 120 includes: the system comprises an access server, an emotion analysis server and a database. The access server is used to provide access services for the terminal 110. The emotion analysis server is used for sentence segmentation, vocabulary extraction, sentence emotion analysis and the like. The number of the emotion analysis servers may be one or more, and when there are multiple emotion analysis servers, there are at least two emotion analysis servers for providing different services, and/or there are at least two emotion analysis servers for providing the same service, for example, providing the same service in a load balancing manner or providing the same service in a manner of a main server and a mirror server, which is not limited in the embodiments of the present application. The database is used for storing word vectors corresponding to the vocabulary meanings, emotion types and the like. The statement to be analyzed is data information which is authorized to be acquired by the user.
The terminal 110 may be generally referred to as one of a plurality of terminals, and the embodiment is only illustrated by the terminal 110. Those skilled in the art will appreciate that the number of terminals described above may be greater or fewer. For example, the number of the terminal may be only one, or several tens or hundreds, or more, and in this case, the program service system further includes other terminals. The number and the type of the terminals are not limited in the embodiments of the present application.
Fig. 2 is a flowchart of a method for analyzing emotion of a sentence according to an embodiment of the present application, as shown in fig. 2. The method comprises the following steps:
201. the server determines a first word vector corresponding to a plurality of words included in the sentence to be analyzed.
In an embodiment of the present application, the server may obtain at least one statement to be analyzed from at least one application for uploading the statement. For any statement to be analyzed, if the statement needs to be participled, such as a Chinese statement, the server can perform participle on the statement to obtain a plurality of vocabularies; if the sentence does not need to be participled, such as an english sentence, the server can directly obtain a plurality of words included in the sentence. The server may store a corresponding relationship between each vocabulary and a word vector, and the server may determine a first word vector corresponding to a plurality of vocabularies according to the corresponding relationship.
In one possible implementation, since the vocabulary may have different meanings in different domains, the server may determine the first word vector corresponding to the plurality of vocabularies according to the basic meaning of the vocabulary and the special meaning of the vocabulary in the target domain. Correspondingly, the steps can be as follows: the server can perform word segmentation on the sentence to be analyzed to obtain a plurality of words. For any vocabulary, the server can obtain a word vector corresponding to the first meaning of the vocabulary and a word vector corresponding to the second meaning of the vocabulary, and the word vector corresponding to the first meaning of the vocabulary and the word vector corresponding to the second meaning of the vocabulary are spliced to obtain the first word vector corresponding to the vocabulary, so that the first word vectors corresponding to a plurality of vocabularies are obtained. The server splices word vectors corresponding to different meanings of the vocabulary, so that the obtained first word vector corresponding to the vocabulary is more appropriate to the meaning of the vocabulary.
Where the first meaning of the word refers to the basic meaning of the word, e.g., "light" the word having the basic meaning of being lean in composition, corresponding to "rich". The second meaning of the vocabulary refers to the special meaning of the vocabulary in the target field, which can be the field of the scene applied by the sentence, such as the travel field, the shopping field, the catering field, the art field, etc. Taking the catering field as an example, the special meaning of the light flavor in the catering field is that the flavor is light and corresponds to the good flavor. The server may store word vectors corresponding to different meanings of each vocabulary, each vocabulary corresponding to at least one word vector.
It should be noted that the application program including the statement upload function may further include other functions according to the application scenario. For example, when the application is an application applied to a shopping scenario, the application may further include functions of searching for a commodity, adding a shopping cart, paying for an order, evaluating an order, and the like. For another example, when the application program is an application program applied to a dining scene, the application program may further include functions of menu display, displaying a position of a food delivery person, dining evaluation, and the like.
202. The server processes the first word vectors corresponding to the vocabularies, the dependency relationships among the vocabularies and the types of the dependency relationships on the vocabularies based on the graph convolution network to obtain second word vectors corresponding to the vocabularies.
In this embodiment, the server may obtain the second word vectors corresponding to the plurality of words based on the graph convolution network. The graph convolution network may be a DREGCN (Dependency Embedded graph convolution network), and a relationship vector corresponding to a Dependency relationship between vocabularies is added to the DREGCN, compared to a conventional GCN (graph convolution network).
Before describing the DREGCN in the embodiments of the present application, we will first describe the conventional GCN. The conventional GCN is a kind of GNN (Graph Neural Networks), and a word vector corresponding to any word at time t can be calculated by the following formula (1).
Where n denotes the number of words, ReLU denotes the activation function, AijRepresenting a relationship matrix, WgThe parameters of the linear transformation are represented,
word vectors representing the jth vocabulary of the l-th level, bgRepresenting the bias parameter.
The DREGCN in the embodiment of the application is improved on the traditional GCN, and the word vector corresponding to any vocabulary at the time t can be obtained by the following formula (2).
Where N denotes the number of words, N denotes the number of types of dependencies between words, ReLU denotes the activation function, AijRepresenting a relationship matrix, Wr representing linear transformation parameters,
representing the jth vocabulary corresponding word vector at the l level, brDenotes a bias parameter, R [ k ]]A relationship vector representing the type of the kth dependency, "; "denotes vector stitching.
In addition, W is as defined aboverAnd brThe method can be obtained through model training, and the model training mode is not limited in the embodiment of the application.
In a possible implementation manner, the server may determine a relationship matrix according to a dependency relationship between a plurality of words, determine a relationship vector corresponding to the dependency relationship according to a type to which the dependency relationship belongs, and then input the first word vector and the relationship matrix corresponding to the plurality of words and the relationship vector corresponding to each dependency relationship into the graph convolution network to obtain the second word vector corresponding to the plurality of words. Accordingly, this step can be realized by the following sub-steps 2021 to 2024:
2021. the server can obtain the dependency relationship among a plurality of vocabularies;
a sentence may comprise a plurality of words, each having a respective part of speech, such as nouns, pronouns, adjectives, prepositions, etc., and each having a respective role, such as subjects, predicates, objects, determinants, etc. The dependency relationship among the words in the sentence can form a multi-layer tree structure, and the words are connected according to the dependency relationship.
For example, take the statement that "coffee is a cheaper item than an expensive ham sandwich" as an example. When the words are divided, the words "coffee", "is", "a", "b", "more expensive", "ham", "sandwich", "cheap" and "things" can be obtained. The dependency relationship between the above words can be seen in fig. 3. FIG. 3 is a diagram illustrating a vocabulary dependency according to an embodiment of the present application. The words with dependency relationship are connected through line segments, such as that 'yes' and 'coffee' have dependency relationship, that 'east-west' and 'yes', 'one', 'cheap', 'ratio' have dependency relationship, that 'ratio' and 'sandwich', that 'east-west' has dependency relationship, that 'sandwich' and 'ratio', 'high-price' and 'ham' have dependency relationship.
2022. And the server determines a relationship matrix according to the dependency relationship, wherein elements in the relationship matrix are used for indicating whether the dependency relationship exists among the vocabularies.
After the server acquires the dependency relationship, a relationship matrix can be constructed according to the dependency relationship, and the number of rows and columns of the relationship matrix is equal to the number of vocabularies. When any two vocabularies have a dependency relationship, the server can set an element which represents whether the dependency relationship exists between the two vocabularies in the relationship matrix as 1; when any two vocabularies do not have a dependency relationship, the server may set an element indicating whether there is a dependency relationship between the two vocabularies to 0 in the relationship matrix.
For example, the above-mentioned sentence "coffee is a cheaper item than an expensive ham sandwich" will be described as an example. After the server obtains the dependency relationship of each vocabulary, the following relationship matrix a may be determined.
The element value of the first row and the first column of the relation matrix A is 1, and the dependency relation between the coffee and the relation matrix A is represented; the element value of the first row and the second column is 1, which indicates that 'coffee' and 'yes' have dependency relationship; the element value of the fourth row and the seventh column is 1, which indicates that the ratio and the sandwich have a dependency relationship; the element value in the ninth column of the fourth row is 1, which means that "ratio" has a dependency relationship with "east-west". And so on.
2023. For any dependency relationship, the server may obtain a relationship vector corresponding to the dependency relationship according to the type to which the dependency relationship belongs.
For any dependency relationship in the above-identified at least one dependency relationship, the server may determine the type to which the dependency relationship belongs, for example, the noun subject that "coffee" is "yes", and accordingly, the predicate that "is" coffee ", the" ham "is a compound word of" sandwich ", which together constitute a" ham sandwich ", the" high-priced "is an adjective of" sandwich ", and the" sandwich "is an object of a preposition of" ratio ". Each type of dependency corresponds to a relationship vector, and the server can determine the relationship vector corresponding to the dependency according to the type of the dependency.
In a possible implementation manner, the server may store a correspondence between the type of the dependency relationship and the relationship vector. The server may obtain the corresponding relationship vector after determining the type to which the dependency relationship belongs according to the corresponding relationship.
It should be noted that the types of dependencies between words are paired, that is, the type of dependency of word a on word B is a, and the type of dependency of word B on word a is B, for example, "coffee" is the noun subject of "yes," and accordingly, "yes" is the predicate of "coffee.
2024. And the server processes the relation matrix and the plurality of relation vectors based on the graph convolution network to obtain second word vectors corresponding to the plurality of words.
After the server obtains the relationship matrix and the plurality of relationship vectors, for any vocabulary in the plurality of vocabularies, a first word vector corresponding to at least one vocabulary having a dependency relationship with the vocabulary can be obtained. The server can input the relationship matrix, the plurality of relationship vectors and the first word vector corresponding to the at least one vocabulary into the graph convolution network to obtain a second word vector corresponding to the vocabulary. Because the dependency relationship among the vocabularies is considered when the graph convolution network is used for processing, the obtained second word vector corresponding to the vocabularies is more closely associated with the second word vectors corresponding to other vocabularies, and the accuracy of the analysis result is improved.
203. The server determines a first information vector corresponding to a plurality of words and a second information vector corresponding to the plurality of words according to a second word vector corresponding to the plurality of words, wherein the first information vector is used for indicating whether any word in the plurality of words is a first target word, and the second information vector is used for indicating the emotion type of each word in the plurality of words.
In this embodiment, after obtaining the second word vectors corresponding to the vocabularies, the server may respectively process the second word vectors corresponding to the vocabularies based on an aspect vocabulary extraction task and an aspect-level emotion classification task to determine the first information vector and the second information vector. The aspect vocabulary extraction task is used for identifying a first target vocabulary and a second target vocabulary which are included in the plurality of vocabularies, wherein the first target vocabulary can be entity vocabularies such as nouns and pronouns, and the second target vocabulary can be emotion vocabulary such as adjectives. The aspect-level emotion classification task can comprise aspect entity recognition, emotion classification and emotion word recognition.
In a possible implementation manner, the aspect vocabulary extracting task may be completed by a second convolutional neural network, and the aspect-level emotion classification task may be completed by a third convolutional neural network. Correspondingly, the implementation manner of this step may be: the server may re-encode the second word vectors corresponding to the plurality of words based on a second convolutional neural network to obtain first information vectors corresponding to the plurality of words. The server can also re-encode second word vectors corresponding to the plurality of words based on a third convolutional neural network to obtain second information vectors corresponding to the plurality of words. The server may splice the first information vector and the second information vector to obtain a third information vector. The server may re-encode the third information vector based on the second convolutional neural network and the third convolutional neural network, respectively, to obtain the updated first information vector and the updated second information vector. After the obtained first information vector and the second information vector are spliced by the server, the first information vector and the second information vector are input into the second convolutional neural network and the third convolutional neural network respectively again for recoding, so that information sharing is realized among different tasks, namely, information of a first target vocabulary obtained by the aspect vocabulary extraction task is shared to the aspect-level emotion classification task, and information of an emotion type of each vocabulary obtained by the aspect-level emotion classification task is shared to the aspect vocabulary extraction task.
In a possible implementation manner, the server may perform information sharing for multiple times, that is, update the first information vector and the second information vector for multiple times. The corresponding steps can be as follows: the server may splice the updated first information vector and the updated second information vector to obtain an updated third information vector. The server may re-encode the updated third information vector based on the second convolutional neural network and the third convolutional neural network, respectively, again. The above splicing and re-encoding process is repeated until a stop update condition is reached. The update stop condition may be that a target iteration number is reached, for example, 2 times, or that the first information vector and the second information vector converge, which is not limited in this embodiment of the application.
In a possible implementation manner, the server can also send the information of the adjective vocabularies obtained in the aspect vocabulary extraction task to the aspect-level emotion analysis task, so that the server updates the second information vector according to the information of the adjective vocabularies when executing the emotion analysis task. Correspondingly, the step of obtaining the updated first information vector and the second information vector by the server may be: the server may obtain interaction transfer information obtained by the second convolutional neural network according to a second word vector corresponding to the plurality of words, where the interaction transfer information is used to indicate a word belonging to a second target word among the plurality of words. The server may re-encode the third information vector based on the third convolutional neural network and the interactive transfer information to obtain an updated second information vector. The server may re-encode the third information vector based on a second convolutional neural network to obtain an updated first information vector. The server shares the information indicating the second target vocabulary in the plurality of vocabularies, namely the information of the adjective vocabulary, to the aspect-level emotion analysis task, so that the second information vector obtained based on the update of the third convolutional neural network is more accurate. Thereby improving the accuracy of the results.
When the server shares information among the different tasks, the information can be shared by the following formula (3). The server may splice the third information vector input last time and the obtained first information vector and second information vector, and then re-encode the spliced third information vector to obtain an updated third information vector.
Wherein the content of the first and second substances,
an information vector f theta representing the i-th vocabulary after the t-th information sharingreThe representation of the re-encoding function is,
an information vector corresponding to the ith vocabulary in the aspect vocabulary extracting task is shown,
and expressing an information vector corresponding to the ith vocabulary in the aspect-level emotion analysis task.
204. And for each vocabulary, when the probability that the vocabulary belongs to the first target vocabulary is determined to be greater than a target probability threshold value according to the first information vector corresponding to the vocabulary by the server, the probability that the vocabulary belongs to each emotion type is determined according to the second information vector corresponding to the vocabulary.
In this embodiment, for each vocabulary, the server may perform full concatenation on the first information vector corresponding to the vocabulary to obtain a probability that the vocabulary belongs to the first target vocabulary or the second target vocabulary. And when the probability that the vocabulary belongs to the first target vocabulary is larger than the target probability threshold, the server performs emotion prediction according to the second information vector corresponding to the vocabulary to obtain the probability that the vocabulary belongs to each emotion type, so that the emotion type of the vocabulary is obtained. Wherein the sum of the probabilities that any vocabulary belongs to each emotion type is 1.
For example, the emotion type of the entity word "coffee" may have three emotion types of "positive", "neutral" and "negative", and in the above sentence, "cheap" means that "coffee" is cheaper than "sandwich", which is an advantage of "coffee", so that the server has a higher probability of belonging to the emotion type of "positive" after emotion prediction on the word "coffee", and thus the emotion type of the entity word "coffee" is "positive".
It should be noted that, the foregoing step 203 and step 204 are one possible implementation manner for the server to determine the emotion type to which at least one first target vocabulary in the plurality of vocabularies belongs based on the second word vectors corresponding to the plurality of vocabularies.
In a possible implementation manner, the server may further obtain a third word vector corresponding to the plurality of words based on the first convolutional neural network, and determine an emotion type to which at least one first target word in the plurality of words belongs based on a fourth word vector obtained by splicing the third word vector and the second word vector. The corresponding implementation steps can be as follows: the server may process the first word vectors corresponding to the plurality of words based on the first convolutional neural network to obtain third word vectors corresponding to the plurality of words, where the third word vectors correspond to the second word vectors one to one. The server can splice the corresponding second word vectors and the corresponding third word vectors to obtain fourth word vectors corresponding to the vocabularies, and the server can determine the emotion type of at least one first target vocabulary in the vocabularies based on the fourth word vectors corresponding to the vocabularies. The step of determining, by the server, the emotion type to which the at least one first target vocabulary belongs based on the fourth word vector corresponding to the plurality of vocabularies may be an implementation manner of determining, by referring to the second word vector corresponding to the plurality of vocabularies, the emotion type to which the at least one first target vocabulary belongs in the plurality of vocabularies, which is not described in detail in this embodiment of the present application.
In addition, in order to more clearly show the architecture of the method for analyzing sentence emotion provided in the embodiment of the present application, reference may be made to fig. 4, where fig. 4 is a schematic diagram of the architecture of the method for analyzing sentence emotion provided in the embodiment of the present application. In fig. 4, a plurality of words obtained by word segmentation of a sentence to be analyzed are input. The coding layer comprises two parts, wherein the first part is to splice the word vector corresponding to the first meaning of each vocabulary and the word vector corresponding to the second meaning of the vocabulary to obtain the first word vector corresponding to the vocabulary. The second part is to process the first word vectors corresponding to the vocabularies, the dependency relationships among the vocabularies and the types of the dependency relationships based on the graph convolution network to obtain second word vectors corresponding to the vocabularies, and of course, the second part can also process the first word vectors corresponding to the vocabularies based on the first convolution neural network to obtain third word vectors corresponding to the vocabularies. The server may splice the corresponding second word vector and the third word vector to obtain a fourth word vector corresponding to the plurality of words. The task-oriented layer comprises a second convolutional neural network
And a third convolutional neural network
AE represents an aspect vocabulary extraction task, and AS represents an aspect-level emotion classification task. The prediction layer is used for predicting the emotion types to which the entity vocabularies belong.
In the embodiment of the application, when the second word vectors corresponding to the vocabularies are determined for the first word vectors corresponding to the vocabularies based on the graph convolution network, the dependency relationships among the vocabularies and the types to which the dependency relationships belong are also considered, so that even if the sentence to be analyzed is long, the relation among the vocabularies can be strengthened through the dependency relationships among the vocabularies, and the accuracy of the analysis result of the emotion types to which the entity vocabularies belong is improved.
Compared with the related technology, the method for analyzing the sentence emotion has the advantages that indexes such as F1 score and accuracy are greatly improved. In order to show the beneficial effects of the application more clearly, comparative experiments are carried out on the application and the related technology. In a comparison experiment, three general data sets are respectively processed by the application and the related technology based on the related technology. The indexes of the comparison experiment are F1 value for identifying the entity, F1 value for identifying the emotion word, the accuracy of aspect-level emotion classification, F1 value for aspect-level emotion classification and F1 value for extracting the entity correctly and classifying the emotion correctly. The results of the comparative experiments can be seen in table 1.
TABLE 1
As can be seen from table 1, the method for analyzing emotion of sentences provided by the present application has significant improvement in each index compared with the related art.
In addition, the application also compares the differences of common GCN, the DREGCN and DREGCN combination information sharing provided by the application and DREGCN combination information sharing and external knowledge. The comparison index used extracts the correct F1 value for the entity and the correct sentiment classification. Wherein the data sets used in the comparison are still the three general data sets described above. The comparative results can be seen in table 2.
TABLE 2
As can be seen from table 2, by using the DREGCN provided in the embodiment of the present application, the F1 values extracted by the entity and classified by the emotion are significantly improved compared to the conventional GCN, and the DREGCN is combined with the information sharing and external knowledge, so that the DREGCN is significantly improved compared to the conventional GCN. Where the external knowledge refers to the vector representation of the words in the embodiments of the application.
Fig. 5 is a block diagram of an apparatus for analyzing emotion of a sentence according to an embodiment of the present application. The apparatus is used for executing the steps when the method for interacting the virtual object is executed, and referring to fig. 5, the apparatus includes: a determination module 501 and a processing module 502.
The system comprises a determining module, a judging module and a judging module, wherein the determining module is used for determining first word vectors corresponding to a plurality of vocabularies included in a sentence to be analyzed;
the processing module is used for processing first word vectors corresponding to the vocabularies, dependency relations among the vocabularies and types of the dependency relations on the vocabularies on the basis of a graph convolution network to obtain second word vectors corresponding to the vocabularies;
and the determining module is further used for determining the emotion type to which at least one first target vocabulary belongs in the vocabularies based on the second word vectors corresponding to the vocabularies.
In a possible implementation manner, the determining module is further configured to perform word segmentation on the sentence to be analyzed to obtain a plurality of words; for any vocabulary, the word vector corresponding to the first meaning of the vocabulary and the word vector corresponding to the second meaning of the vocabulary are spliced to obtain the first word vector corresponding to the vocabulary.
In another possible implementation manner, the processing module is further configured to obtain a dependency relationship between a plurality of vocabularies; determining a relation matrix according to the dependency relationship, wherein elements in the relation matrix are used for indicating whether the dependency relationship exists among all the vocabularies; for any dependency relationship, acquiring a relationship vector corresponding to the dependency relationship according to the type of the dependency relationship; and processing the relation matrix and the plurality of relation vectors based on the graph convolution network to obtain second word vectors corresponding to the plurality of words.
In another possible implementation manner, the processing module is further configured to, for any vocabulary in the plurality of vocabularies, obtain a first word vector corresponding to at least one vocabulary having a dependency relationship with the vocabulary; and inputting the relation matrix, the plurality of relation vectors and the first word vector corresponding to at least one vocabulary into a graph convolution network to obtain a second word vector corresponding to the vocabulary.
In another possible implementation manner, the determining module is further configured to process the first word vectors corresponding to the multiple vocabularies based on the first convolutional neural network to obtain third word vectors corresponding to the multiple vocabularies, where the third word vectors correspond to the second word vectors one to one; splicing the corresponding second word vector and the third word vector to obtain a fourth word vector corresponding to a plurality of vocabularies; and determining the emotion type of at least one first target vocabulary in the plurality of vocabularies based on the fourth word vectors corresponding to the plurality of vocabularies.
In another possible implementation manner, the determining module is further configured to determine, according to a second word vector corresponding to a plurality of words, a first information vector corresponding to the plurality of words and a second information vector corresponding to the plurality of words, where the first information vector is used to indicate whether any word in the plurality of words is a first target word, and the second information vector is used to indicate an emotion type to which each word in the plurality of words belongs; and for each vocabulary, when the probability that the vocabulary belongs to the first target vocabulary is determined to be greater than the target probability threshold value according to the first information vector corresponding to the vocabulary, the probability that the vocabulary belongs to each emotion type is determined according to the second information vector corresponding to the vocabulary.
In another possible implementation manner, the determining module is further configured to re-encode second word vectors corresponding to the multiple words based on a second convolutional neural network to obtain first information vectors corresponding to the multiple words; recoding second word vectors corresponding to the vocabularies on the basis of a third convolutional neural network to obtain second information vectors corresponding to the vocabularies; splicing the first information vector and the second information vector to obtain a third information vector; and respectively recoding the third information vector based on the second convolutional neural network and the third convolutional neural network to obtain the updated first information vector and the updated second information vector.
In another possible implementation manner, the apparatus further includes:
the splicing module is used for splicing the updated first information vector and the updated second information vector to obtain an updated third information vector;
the encoding module is used for respectively re-encoding the updated third information vector again based on the second convolutional neural network and the third convolutional neural network;
and the updating module is used for repeating the splicing and recoding processes until the updating stopping condition is reached.
In another possible implementation manner, the determining module is further configured to obtain interactive transfer information obtained by the second convolutional neural network according to a second word vector corresponding to the plurality of words, where the interactive transfer information is used to indicate a word belonging to a second target word in the plurality of words; recoding the third information vector based on the third convolutional neural network and the interactive transfer information to obtain an updated second information vector; and recoding the third information vector based on the second convolutional neural network to obtain the updated first information vector.
In the embodiment of the application, when the second word vectors corresponding to the vocabularies are determined for the first word vectors corresponding to the vocabularies based on the graph convolution network, the dependency relationships among the vocabularies and the types to which the dependency relationships belong are also considered, so that even if the sentence to be analyzed is long, the relation among the vocabularies can be strengthened through the dependency relationships among the vocabularies, and the accuracy of the analysis result of the emotion types to which the vocabularies belong is improved.
Fig. 6 is a schematic structural diagram of a server 600 according to an embodiment of the present application. The server 600 may have a relatively large difference due to different configurations or performances, and may include one or more processors (CPUs) 601 and one or more memories 602, where the memory 602 stores at least one instruction, and the at least one instruction is loaded and executed by the processor 601 to implement the methods provided by the method embodiments. Of course, the server may also have components such as a wired or wireless network interface, a keyboard, and an input/output interface, so as to perform input/output, and the server may also include other components for implementing the functions of the device, which are not described herein again.
The embodiment of the application also provides a storage medium, which is applied to a server, and at least one program code is stored in the storage medium and used for being executed by a processor and realizing the method for analyzing the sentence emotion in the embodiment of the application.
It will be understood by those skilled in the art that all or part of the steps of implementing the above embodiments may be implemented by hardware, or may be implemented by a program instructing relevant hardware, and the program may be stored in a storage medium, such as a read-only memory, a magnetic disk or an optical disk.
The present application is intended to cover various modifications, alternatives, and equivalents, which may be included within the spirit and scope of the present application.
Claims (12)
1. A method for sentiment analysis of sentences, the method comprising:
determining a first word vector corresponding to a plurality of words included in a sentence to be analyzed;
processing first word vectors corresponding to the vocabularies, dependency relations among the vocabularies and types of the dependency relations on the vocabularies on the basis of a graph convolution network to obtain second word vectors corresponding to the vocabularies;
and determining the emotion type to which at least one first target vocabulary in the vocabularies belongs based on the second word vectors corresponding to the vocabularies.
2. The method of claim 1, wherein determining a first word vector corresponding to a plurality of words included in the sentence to be analyzed comprises:
performing word segmentation on a sentence to be analyzed to obtain a plurality of words;
for any vocabulary, splicing word vectors corresponding to the first meanings of the vocabulary and word vectors corresponding to the second meanings of the vocabulary to obtain the first word vectors corresponding to the vocabulary.
3. The method of claim 1, wherein the processing the first word vectors corresponding to the vocabularies, the dependencies among the vocabularies, and the types to which the dependencies belong based on the graph convolution network to obtain the second word vectors corresponding to the vocabularies comprises:
acquiring a dependency relationship among the vocabularies;
determining a relation matrix according to the dependency relation, wherein elements in the relation matrix are used for indicating whether dependency relation exists among all vocabularies;
for any dependency relationship, acquiring a relationship vector corresponding to the dependency relationship according to the type of the dependency relationship;
and processing the relation matrix and the plurality of relation vectors based on the graph convolution network to obtain second word vectors corresponding to the plurality of words.
4. The method of claim 3, wherein the processing the relationship matrix and the plurality of relationship vectors based on the graph convolution network to obtain second word vectors corresponding to the plurality of words comprises:
for any vocabulary in the plurality of vocabularies, acquiring a first word vector corresponding to at least one vocabulary having a dependency relationship with the vocabulary;
and inputting the relation matrix, the plurality of relation vectors and the first word vector corresponding to the at least one vocabulary into the graph convolution network to obtain a second word vector corresponding to the vocabulary.
5. The method of claim 1, wherein determining an emotional type to which at least a first target vocabulary of the plurality of vocabularies belongs based on a second word vector corresponding to the plurality of vocabularies comprises:
processing first word vectors corresponding to the vocabularies on the basis of a first convolution neural network to obtain third word vectors corresponding to the vocabularies, wherein the third word vectors correspond to the second word vectors one by one;
splicing the corresponding second word vector and the third word vector to obtain a fourth word vector corresponding to the plurality of words;
and determining the emotion type to which at least one first target vocabulary in the vocabularies belongs based on the fourth word vectors corresponding to the vocabularies.
6. The method of claim 1, wherein determining an emotional type to which at least a first target vocabulary of the plurality of vocabularies belongs based on a second word vector corresponding to the plurality of vocabularies comprises:
determining a first information vector corresponding to the vocabularies and a second information vector corresponding to the vocabularies according to second word vectors corresponding to the vocabularies, wherein the first information vector is used for indicating whether any vocabulary in the vocabularies is a first target vocabulary or not, and the second information vector is used for indicating the emotion types of each vocabulary in the vocabularies;
and for each vocabulary, when the probability that the vocabulary belongs to a first target vocabulary is determined to be greater than a target probability threshold value according to a first information vector corresponding to the vocabulary, determining the probability that the vocabulary belongs to each emotion type according to a second information vector corresponding to the vocabulary.
7. The method of claim 6, wherein determining the first information vector corresponding to the plurality of words and the second information vector corresponding to the plurality of words from the second word vector corresponding to the plurality of words comprises:
recoding second word vectors corresponding to the vocabularies on the basis of a second convolutional neural network to obtain first information vectors corresponding to the vocabularies;
recoding second word vectors corresponding to the vocabularies on the basis of a third convolutional neural network to obtain second information vectors corresponding to the vocabularies;
splicing the first information vector and the second information vector to obtain a third information vector;
and respectively recoding the third information vector based on the second convolutional neural network and the third convolutional neural network to obtain the updated first information vector and second information vector.
8. The method of claim 7, wherein after obtaining the updated first information vector and the second information vector, the method further comprises:
splicing the updated first information vector and the updated second information vector to obtain an updated third information vector;
re-encoding the updated third information vector based on the second convolutional neural network and the third convolutional neural network again respectively;
the above splicing and re-encoding process is repeated until a stop update condition is reached.
9. The method of claim 7, wherein re-encoding the third information vector based on the second convolutional neural network and the third convolutional neural network, respectively, to obtain updated first information vector and second information vector, comprises:
acquiring interactive transmission information obtained by the second convolutional neural network according to second word vectors corresponding to the plurality of words, wherein the interactive transmission information is used for indicating words belonging to a second target word in the plurality of words;
recoding the third information vector based on the third convolutional neural network and the interactive transfer information to obtain an updated second information vector;
and recoding the third information vector based on the second convolutional neural network to obtain an updated first information vector.
10. An apparatus for emotion analysis of sentences, the apparatus comprising:
the system comprises a determining module, a judging module and a judging module, wherein the determining module is used for determining first word vectors corresponding to a plurality of vocabularies included in a sentence to be analyzed;
the processing module is used for processing the first word vectors corresponding to the vocabularies, the dependency relationships among the vocabularies and the types of the dependency relationships on the vocabularies on the basis of a graph convolution network to obtain second word vectors corresponding to the vocabularies;
the determining module is further configured to determine an emotion type to which at least one first target vocabulary in the multiple vocabularies belongs based on the second word vectors corresponding to the multiple vocabularies.
11. A server, characterized in that the server comprises a processor and a memory, the memory is used for storing at least one program code, the at least one program code is loaded by the processor and executes the method for sentence emotion analysis according to any of claims 1 to 9.
12. A storage medium for storing at least one program code for execution by a processor and implementing a method for sentiment analysis of sentences as claimed in any one of claims 1 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911147402.3A CN110929528B (en) | 2019-11-21 | 2019-11-21 | Method, device, server and storage medium for analyzing emotion of sentence |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911147402.3A CN110929528B (en) | 2019-11-21 | 2019-11-21 | Method, device, server and storage medium for analyzing emotion of sentence |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110929528A true CN110929528A (en) | 2020-03-27 |
| CN110929528B CN110929528B (en) | 2023-09-05 |
Family
ID=69851465
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911147402.3A Active CN110929528B (en) | 2019-11-21 | 2019-11-21 | Method, device, server and storage medium for analyzing emotion of sentence |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110929528B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111861546A (en) * | 2020-06-28 | 2020-10-30 | 浪潮(北京)电子信息产业有限公司 | Comment quality judgment method, device, equipment and storage medium |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103365867A (en) * | 2012-03-29 | 2013-10-23 | 腾讯科技（深圳）有限公司 | Method and device for emotion analysis of user evaluation |
| CN103455562A (en) * | 2013-08-13 | 2013-12-18 | 西安建筑科技大学 | Text orientation analysis method and product review orientation discriminator on basis of same |
| US20140365207A1 (en) * | 2013-06-06 | 2014-12-11 | Xerox Corporation | Method and system for classifying reviewers' comments and recommending related actions in idea-generating social media platforms |
| US20170192955A1 (en) * | 2015-12-30 | 2017-07-06 | Nice-Systems Ltd. | System and method for sentiment lexicon expansion |
| CN110377740A (en) * | 2019-07-22 | 2019-10-25 | 腾讯科技（深圳）有限公司 | Feeling polarities analysis method, device, electronic equipment and storage medium |
| WO2019214145A1 (en) * | 2018-05-10 | 2019-11-14 | 平安科技（深圳）有限公司 | Text sentiment analyzing method, apparatus and storage medium |
| CN110472003A (en) * | 2019-08-08 | 2019-11-19 | 东北大学 | Social networks text emotion fine grit classification method based on figure convolutional network |
-
2019
- 2019-11-21 CN CN201911147402.3A patent/CN110929528B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103365867A (en) * | 2012-03-29 | 2013-10-23 | 腾讯科技（深圳）有限公司 | Method and device for emotion analysis of user evaluation |
| US20140365207A1 (en) * | 2013-06-06 | 2014-12-11 | Xerox Corporation | Method and system for classifying reviewers' comments and recommending related actions in idea-generating social media platforms |
| CN103455562A (en) * | 2013-08-13 | 2013-12-18 | 西安建筑科技大学 | Text orientation analysis method and product review orientation discriminator on basis of same |
| US20170192955A1 (en) * | 2015-12-30 | 2017-07-06 | Nice-Systems Ltd. | System and method for sentiment lexicon expansion |
| WO2019214145A1 (en) * | 2018-05-10 | 2019-11-14 | 平安科技（深圳）有限公司 | Text sentiment analyzing method, apparatus and storage medium |
| CN110377740A (en) * | 2019-07-22 | 2019-10-25 | 腾讯科技（深圳）有限公司 | Feeling polarities analysis method, device, electronic equipment and storage medium |
| CN110472003A (en) * | 2019-08-08 | 2019-11-19 | 东北大学 | Social networks text emotion fine grit classification method based on figure convolutional network |
Non-Patent Citations (1)
| Title |
|---|
| 喻涛;罗可;: "利用动态多池卷积神经网络的情感分析模型", 计算机科学与探索, no. 07 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111861546A (en) * | 2020-06-28 | 2020-10-30 | 浪潮(北京)电子信息产业有限公司 | Comment quality judgment method, device, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110929528B (en) | 2023-09-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11334635B2 (en) | Domain specific natural language understanding of customer intent in self-help | |
| Bhatia | Data mining and data warehousing: principles and practical techniques | |
| US20240080531A1 (en) | Profiling media characters | |
| Mahajan et al. | Parameter estimation in marketing models in the presence of multicollinearity: An application of ridge regression | |
| US20130159348A1 (en) | Computer-Implemented Systems and Methods for Taxonomy Development | |
| US10360219B2 (en) | Applying level of permanence to statements to influence confidence ranking | |
| CN110110233B (en) | Information processing method, device, medium and computing equipment | |
| US9251527B2 (en) | System and method for providing personalized recommendations | |
| US20210390609A1 (en) | System and method for e-commerce recommendations | |
| CN110362663B (en) | Adaptive multi-perceptual similarity detection and analysis | |
| US11144569B2 (en) | Operations to transform dataset to intent | |
| CN115917535A (en) | Recommendation model training method, recommendation device and computer readable medium | |
| US11874798B2 (en) | Smart dataset collection system | |
| US20210397634A1 (en) | Automated processing of unstructured text data in paired data fields of a document | |
| US20200175390A1 (en) | Word embedding model parameter advisor | |
| CN111754278A (en) | Article recommendation method and device, computer storage medium and electronic equipment | |
| CN111931055B (en) | Object recommendation method, object recommendation device and electronic equipment | |
| US11797281B2 (en) | Multi-language source code search engine | |
| CN113641707A (en) | Knowledge graph disambiguation method, device, equipment and storage medium | |
| US20190295110A1 (en) | Performance analytics system for scripted media | |
| CN110929528B (en) | Method, device, server and storage medium for analyzing emotion of sentence | |
| CN113032676A (en) | Recommendation method and system based on micro-feedback | |
| CN116756281A (en) | Knowledge question-answering method, device, equipment and medium | |
| CN115080856A (en) | Recommendation method and device and training method and device of recommendation model | |
| CN113094584A (en) | Method and device for determining recommended learning resources |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| REG | Reference to a national code |
Ref country code: HK Ref legal event code: DE Ref document number: 40021093 Country of ref document: HK |
|
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |