CN111125360B - Emotion analysis method and device in game field and model training method and device thereof - Google Patents
Emotion analysis method and device in game field and model training method and device thereof Download PDFInfo
- Publication number
- CN111125360B CN111125360B CN201911322985.9A CN201911322985A CN111125360B CN 111125360 B CN111125360 B CN 111125360B CN 201911322985 A CN201911322985 A CN 201911322985A CN 111125360 B CN111125360 B CN 111125360B
- Authority
- CN
- China
- Prior art keywords
- emotion
- word
- training
- text
- category
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Machine Translation (AREA)
Abstract
The application provides an emotion analysis method and device in the game field and a model training method and device thereof, relates to the technical field of language processing, and solves the technical problem of lower accuracy of emotion analysis results. The method comprises the following steps: determining a first object word segmentation belonging to a first category in word segmentation of a first text of the game field to be analyzed according to a first number of categories of the preset game field; replacing the first object word segmentation in the first text with the category word segmentation of the first category to obtain a second text; and applying a pre-trained emotion analysis model to predict emotion according to the second text, and obtaining emotion tendencies corresponding to the first category.
Description
Technical Field
The application relates to the technical field of language processing, in particular to an emotion analysis method and device and a model training method and device thereof in the field of games.
Background
Text emotion analysis is the process of analyzing, processing, generalizing and reasoning subjective text with emotion colors. By performing emotion analysis on the text, particularly on specific objects in the text, people can be helped to more effectively sort out emotion tendencies based on the related text.
At present, in the existing emotion analysis process for a specific object in a text, the text to be analyzed and the specific object in the text are used as input of a model, adjectives near the specific object are used as basis for emotion judgment of the specific object, and emotion tendencies of the specific object in the text are obtained.
However, this approach can result in missing new words outside of some specific objects, especially in the field of games, where some new words are often present. Therefore, the current emotion analysis method is easy to influence the accuracy of emotion analysis results due to inaccurate analysis of new words.
Disclosure of Invention
The application aims to provide an emotion analysis method and device in the game field and a model training method and device thereof, so as to solve the technical problem of lower accuracy of emotion analysis results.
In a first aspect, an embodiment of the present application provides an emotion analysis method in a game field, including:
determining a first object word segmentation belonging to a first category in word segmentation of a first text of the game field to be analyzed according to a first number of categories of the preset game field;
replacing the first object word segmentation in the first text with the category word segmentation of the first category to obtain a second text;
And applying a pre-trained emotion analysis model to predict emotion according to the second text, and obtaining emotion tendencies corresponding to the first category.
In one possible implementation, each of the first number of categories corresponds to a set of object segmentation clusters; according to a preset first number of categories of the game field, determining a first object word segmentation belonging to the first category in word segmentation of a first text of the game field to be analyzed, wherein the step comprises the following steps:
word segmentation is carried out on the first text;
matching the word segments in the first text in all object word segment clustering sets corresponding to the first number of categories;
if the first object segmentation in the first text is successfully matched in the first object segmentation cluster set, determining that the first object segmentation belongs to a first category corresponding to the first object segmentation cluster set.
In one possible implementation, the method further includes:
if the word segmentation in the first text is not successfully matched in all the object word segmentation clustering sets;
clustering according to the word segmentation clustering set of all the objects and the word segmentation in the first text;
and if the first object segmentation word cluster in the first text is determined to be in the first object segmentation word cluster set, determining that the first object segmentation word belongs to a first category corresponding to the first object segmentation word cluster set.
In one possible implementation, before the step of determining, according to a preset first number of categories of the game field, a first object word segment belonging to the first category in the word segment of the first text of the game field to be analyzed, the method further includes:
preprocessing a first original corpus in the game field based on word frequency and word parts to obtain an object word segmentation set;
clustering the object word segmentation set according to a first number of categories, and determining an object word segmentation cluster set corresponding to each category in the first number of categories.
In one possible implementation, the first number of categories includes: play, activity, occupation, important roles, props, player behavior, version, and teams.
In one possible implementation, the pre-trained emotion analysis model includes a pre-trained language model and a pre-trained classification model; applying a pre-trained emotion analysis model, and carrying out emotion prediction according to the second text to obtain emotion tendencies corresponding to the first category, wherein the method comprises the following steps:
converting the second text into sentence vectors by applying a pre-trained language model;
applying a pre-trained classification model, and determining emotion tendencies corresponding to the first category according to the sentence vectors; wherein the emotional tendency includes positive, other, and negative.
In a second aspect, there is provided an emotion analysis model training method in the field of games, including:
determining a training text sample set, wherein each training text sample in the training text sample set corresponds to an emotion tendency label, and each training text sample comprises a category word segmentation;
and training the emotion analysis model according to the training text sample set.
In one possible implementation, the step of determining a set of training text samples includes:
filtering a second original corpus of the game field according to object segmentation corresponding to a first number of categories of the preset game field, and determining an initial text sample set, wherein each initial text sample in the initial text sample set comprises the object segmentation;
replacing the object word of each initial text sample in the initial text sample set with the category word corresponding to the object word to obtain an intermediate text sample set;
and determining a training text sample set according to the intermediate text sample set and the emotion tendency labels of each intermediate text sample.
In one possible implementation, each of the first number of categories corresponds to a set of object segmentation clusters; the method comprises the steps of filtering a second original corpus of the game field according to object word segmentation corresponding to a first number of categories of the preset game field, and determining an initial text sample set, wherein the steps comprise:
Word segmentation is carried out on each original text in the second original corpus;
deleting the original text which does not exist in the second original corpus and is contained in all the object word segmentation cluster sets and does not exist the word segments clustered into all the object word segmentation cluster sets, and obtaining an initial text sample set.
In one possible implementation, the training the emotion analysis model according to the training text sample set includes:
sequentially selecting a current training text sample from the training text sample set, and performing the following steps until the analysis result of the emotion analysis model reaches the expectation, and outputting a pre-trained emotion analysis model; the pre-trained emotion analysis model comprises a pre-trained language model and a pre-trained classification model;
inputting the current training text sample and the current class word of the current training text sample into the pre-trained language model, and outputting a first sentence vector of the current training text sample and a second sentence vector of the current class word;
inputting the first sentence vector and the second sentence vector into a classification model to obtain a preliminary emotion tendency;
And optimizing the pre-trained language model and the classification model based on the preliminary emotion tendency and the current emotion tendency label corresponding to the current training text sample by taking the difference between the preliminary emotion tendency and the current emotion tendency label as a target, and taking the optimized language model and classification model as a new emotion analysis model to continue training.
In one possible implementation, the emotional tendency tags include positive, other, and negative.
In a third aspect, an embodiment of the present application further provides an emotion analysis device in a game field, including:
the determining module is used for determining first object word segmentation belonging to a first category in word segmentation of a first text of the game field to be analyzed according to a first number of categories of the preset game field;
the replacing module is used for replacing the first object word segmentation in the first text with the category word segmentation of the first category to obtain a second text;
and the prediction module is used for applying a pre-trained emotion analysis model, and predicting emotion according to the second text to obtain emotion tendencies corresponding to the first category.
In a fourth aspect, an embodiment of the present application further provides an emotion analysis model training device in the field of games, including:
The system comprises a determining unit, a judging unit and a judging unit, wherein the determining unit is used for determining a training text sample set, each training text sample in the training text sample set corresponds to an emotion tendency label, and each training text sample comprises a category word;
and the training unit is used for training the emotion analysis model according to the training text sample set.
In a fifth aspect, an embodiment of the present application further provides a computer device, including a memory, and a processor, where the memory stores a computer program that can be executed by the processor, and the processor executes the method according to the first aspect or the second aspect.
In a sixth aspect, embodiments of the present application further provide a computer readable storage medium storing machine executable instructions which, when invoked and executed by a processor, cause the processor to perform the method of the first or second aspects described above.
The embodiment of the application has the following beneficial effects:
according to the emotion analysis method and device for the game field and the model training method and device thereof, which are provided by the embodiment of the application, according to the preset first number of categories in the game field, a first object word belonging to the first category in word segmentation of a first text in the game field to be analyzed can be determined, then the first object word in the first text is replaced by the category word of the first category to obtain a second text, finally, a pre-trained emotion analysis model is applied, emotion prediction is performed according to the second text to obtain emotion tendency corresponding to the first category, emotion analysis is performed on the category word to which the object word belongs by determining the category word to which the object word belongs, and emotion analysis can be performed when a new object word which does not appear is faced, so that universality is improved by better utilizing existing data, and emotion analysis accuracy of the object word in the text is improved.
In order to make the above objects, features and advantages of the present application more comprehensible, preferred embodiments accompanied with figures are described in detail below.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings that are needed in the description of the embodiments or the prior art will be briefly described, and it is obvious that the drawings in the description below are some embodiments of the present application, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flowchart of an emotion analysis method in a game field provided by an embodiment of the present application;
FIG. 2 is a flowchart of an emotion analysis model training method in the game field provided by an embodiment of the present application;
FIG. 3 is another flowchart of an emotion analysis model training method in the game field provided by an embodiment of the present application;
FIG. 4 is a schematic diagram of an emotion analysis device in the game field according to an embodiment of the present application;
FIG. 5 is a schematic structural diagram of an emotion analysis model training device in the game field according to an embodiment of the present application;
Fig. 6 is a schematic structural diagram illustrating a computer device according to an embodiment of the present application.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present application more apparent, the technical solutions of the present application will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
The terms "comprising" and "having" and any variations thereof, as used in the embodiments of the present application, are intended to cover non-exclusive inclusion. For example, a process, method, system, article, or apparatus that comprises a list of steps or elements is not limited to only those listed but may optionally include other steps or elements not listed or inherent to such process, method, article, or apparatus.
The rapid development of the Internet brings increasingly rich web text comment data, and by means of emotion classification of the web text comment data, related people can be helped to better sort out emotion tendencies based on related texts. Emotion analysis is one of the main classification analysis tasks of natural language processing (Natural Language Process, NLP). And the emotion analysis task encodes the network comment text, and then classifies the comment text according to the marking data standard by the encoding result. The emotion analysis task can carry out text emotion classification and identification through a classification algorithm, and can also carry out small-amplitude fine adjustment and classification and identification on the classification task according to a pre-trained language model (Bidirectional Encoder Representation from Transformers, bert).
The emotion analysis result of the comment data needs to be higher in accuracy, and meanwhile, emotion tendencies of specific objects in the comment data need to be identified in a finer granularity emotion analysis mode. In the existing fine granularity emotion analysis mode, adjectives near a specific object are only considered as the basis for judging the emotion of the specific object, the specific adjectives are directly searched according to the part of speech in the network text by considering the specificity of the existing network text, most newly-appearing network adjectives are easy to miss, for example, a 'you are true siblings', and an existing word separator cannot obtain the siblings in the current context.
At present, the prior art using a neural network method is mostly based on word segmentation, and in consideration of the specificity of network texts, new words are very easy to appear, and the problem of low final accuracy caused by inaccurate word segmentation is easily caused. For example, "play the game, i feel a panic" and when encountering a new word "a panic" the word segmenter cannot recognize it as a complete word, resulting in misjudging the emotional tendency of the game subject.
And carrying out fine granularity emotion analysis on word vectors and specific objects of comment texts after modeling through a neural network, wherein the generalization effect is poor when the word vectors and the specific objects are faced with the specific objects which are not in the training corpus. Considering the divergence of the network general text, training is carried out through limited corpus, so that the accuracy of fine granularity emotion analysis is not high easily.
Therefore, the current training by using the current neural network method is mostly only aimed at specific objects in the general text, and when the model faces the specific objects which do not appear in the training set, the model has the characteristics of low accuracy and low generalization capability because different analysis objects can cause different language styles and structures of the network text.
Furthermore, the existing neural network method cannot solve the problem of long-distance dependence of adjectives and specific objects. Only adjectives near a specific object are considered as emotion judgment of the specific object in a regular mode, and only adjectives near the specific object are considered, so that the problem of long-distance dependence of adjectives and the specific object cannot be solved, for example, the restaurant meal service is still to be improved. When the object is "service", the latest adjective is displayed as "good" and is determined as positive, but actually true emotion tendency for the object to be "service" is negative.
Therefore, the method only considers adjectives near the specific object as the basis for judging the emotion of the specific object, and when the adjectives are positive adjectives, the adjectives represent positive emotion to the specific object, and when the adjectives are negative adjectives, the adjectives represent negative emotion to the specific object. This results in a lower accuracy of the emotion analysis prediction results.
The embodiment of the application provides an emotion analysis method and device in the game field and a model training method and device thereof, and the technical problem of lower accuracy of emotion analysis results can be solved by the method.
Embodiments of the present application are further described below with reference to the accompanying drawings.
Fig. 1 is a schematic flow chart of an emotion analysis method in the game field according to an embodiment of the present application. As shown in fig. 1, the method includes:
s110, determining a first object word segment belonging to a first category in word segments of a first text of the game field to be analyzed according to a first number of categories of the preset game field.
The text may be divided into object words and attribute words, for example, the object words may be "a teacher" and the attribute words may be "a playable".
The first text of the game field to be analyzed may be collected web text related to a plurality of games. Such as comment text for posts and replies related to games, such as bars, microblogs, game officer web forums, etc.
The object segmentation may be the segmentation of the object of interest commonly found in the web text obtained by processing the web text by a correlation algorithm. In the embodiment of the application, the common object segmentation words can be divided into a first number of categories so as to adapt to different objects of a plurality of game specific subdivisions, and the universality of the emotion analysis method provided by the embodiment of the application is enhanced.
S120, replacing the first object word in the first text with the category word of the first category to obtain a second text.
The category word is a word indicating the type of the object word, and may be, for example, "play", "activity", "occupation", "important role", "prop", "player behavior", "version", "team".
As another example, the object words that may be included in the category word "occupation" may be object words belonging to the occupation such as "mr", "fighter", "morgan", "iron coat" and "blood river".
S130, applying a pre-trained emotion analysis model, and carrying out emotion prediction according to the second text to obtain emotion tendencies corresponding to the first category.
In this step, the second text is input into a pre-trained emotion analysis model, and the relative emotion tendencies of the first text for the category of the object segmentation can be output.
Because the semantic style and sentence pattern similarity of the network text where the object word belonging to one class word is located are larger, in the step, the targeted emotion analysis can be directly carried out on the class word to which the object word belongs, so that the emotion analysis method provided by the embodiment of the application has stronger universality.
In the existing process of carrying out fine granularity emotion analysis on a text by utilizing a neural network, the text and a specific object are taken as the input of a model, adjectives near the specific object are taken as the basis of emotion judgment of the specific object, and emotion tendencies of the specific object are obtained after relevant weight processing is carried out according to a downstream network. However, the method is only based on the inherent object segmentation, only considers the inherent specific object, directly searches the specific adjective according to the specific object in the web text, and easily omits most newly appeared web vocabularies. However, due to the divergence of the web text, new words outside the inherent specific objects are very easy to appear, and the new words of the objects which do not appear in the inherent specific objects are easy to cause lower accuracy of emotion analysis results because of inaccurate analysis of the new words.
In the embodiment of the application, unlike the prior art that only a specific object is analyzed, the class word to which each object word belongs is determined, and the related fine-grained emotion analysis is performed on the class word to which the object word belongs, so that the fine-grained emotion analysis can be performed when a new object word which does not appear is faced, the universality is improved by better utilizing the existing data, and the accuracy and the robustness of the emotion analysis on the object word are improved.
The above steps are described in detail below.
In some embodiments, it may be determined, through a matching process, whether the web text to be analyzed includes the object segmentation in the first number of categories. As one example, each of the first number of categories corresponds to a set of object segmentation clusters; the step S110 may include the steps of:
and a, word segmentation is carried out on the first text.
And b, matching the word segmentation in the first text in all object word segmentation clustering sets corresponding to the first number of categories.
And c, if the first object segmentation in the first text is successfully matched in the first object segmentation clustering set, determining that the first object segmentation belongs to a first category corresponding to the first object segmentation clustering set.
In the step b, the matching process may be to match each object word or match all the words.
In the embodiment of the application, firstly, screening and extracting a specific object from the network text to be analyzed, and checking whether the network text contains object segmentation in the object segmentation cluster set or not through a matching process. If the specific object is contained, determining the category corresponding to the object word segmentation cluster set to which the object word segmentation belongs, replacing the object word segmentation in the network text with the corresponding category word segmentation, and taking the replaced text as the input of a subsequent emotion analysis model.
Through a pre-matching process, whether the network text to be analyzed contains the object word in the first number of categories can be judged, so that the category to which the object word belongs can be more effectively determined under the condition of containing the object word.
In some embodiments, the object segmentation need not be preset to a certain number of specific objects, but need only be satisfied in a preset category. As an example, the method may further comprise the steps of:
and e, if the word segmentation in the first text is not successfully matched in all the object word segmentation clustering sets, clustering according to all the object word segmentation clustering sets and the word segmentation in the first text.
And f, if the first object segmentation word in the first text is determined to be clustered into a first object segmentation word cluster set, determining that the first object segmentation word belongs to a first category corresponding to the first object segmentation word cluster set.
If the text to be analyzed does not contain the object word in the first number of categories, the category to which the word in the text belongs can be determined in the text to be analyzed through a clustering process, so that the object word is not limited to the inherent object word clustering set, the object word in the first number of categories can be wider, and the universality of the emotion analysis method in the game field can be met.
In some embodiments, the collected object segmentation may be classified in advance to obtain a cluster set of object segmentation clusters corresponding to each category. As an example, before the step S110, the method may further include the steps of:
and g, preprocessing the first original corpus in the game field based on word frequency and word parts to obtain an object word segmentation set.
And h, clustering the object word segmentation set according to a first number of categories, and determining an object word segmentation cluster set corresponding to each category in the first number of categories.
In practical application, data of all channels related to the game in the whole network can be collected first, for example, comment objects common in the game field are collected, so that related abstract cluster analysis is performed on the comment objects.
The method is characterized in that byte pair encoder bytes are utilized to analyze the collected related data by an encoding algorithm, so that a plurality of high-frequency words are obtained, the parts of speech of the words are utilized, nouns and some new words are reserved, and preliminary cleaning is carried out. And then screening, wherein the screened objects can determine a specific object set which is frequently concerned by the comments in the game field. After the specific object set is obtained, the set of common specific objects in the game field is abstracted into eight categories, namely playing methods, activities, professions, important roles, props, player behaviors, versions and warriors by utilizing a clustering algorithm. The categories of a particular object, such as profession, that is subdivided below each category include: the words of a teacher, a warrior, a morale, an iron coat, a blood river and the like.
Before emotion analysis is carried out on a first text to be analyzed, object segmentation words in a first original corpus are collected in advance, so that a first number of categories and object segmentation word clustering sets corresponding to each category are obtained, the object segmentation word clustering sets can be directly utilized in the emotion analysis process, and the categories of segmentation words in the first text to be analyzed can be rapidly determined, so that the efficiency of the emotion analysis process is improved.
In some embodiments, the content of the category word segment of the game field may be determined. Based thereon, the first number of categories includes: play, activity, occupation, important roles, props, player behavior, version, and teams.
Through the categories in the game fields of play, activities, professions, important roles, props, player behaviors, versions, teams and the like, the emotion analysis process can be more targeted, so that more accurate emotion tendencies can be analyzed aiming at the categories.
In some embodiments, emotion analysis may be performed using a language model and a classification model, respectively. As one example, the pre-trained emotion analysis model includes a pre-trained language model and a pre-trained classification model; the step S130 may include the steps of:
And i, converting the second text into sentence vectors by applying a pre-trained language model.
And j, applying a pre-trained classification model, and determining emotion tendencies corresponding to the first category according to the sentence vectors.
Among these emotional trends include positive, other, and negative.
In practical application, the second text obtained in step S120 is used as an input of a pre-trained language model, sentence vectors are output, analysis and prediction of fine granularity emotion tendencies are performed by using the sentence vectors as an input of a pre-trained classification model, forward emotion tendencies, other emotion tendencies or negative emotion tendencies and probabilities thereof are output, and emotion with the highest probability of the three emotion tendencies is taken as emotion of the first text to be returned to the user side.
Through the pre-trained language model and the pre-trained classification model, the analysis and prediction process of fine-grained emotion tendencies can be performed more systematically, so that more accurate emotion tendency analysis results can be obtained.
Fig. 2 is a schematic flow chart of an emotion analysis model training method in the game field. As shown in fig. 2, the method includes:
s210, determining a training text sample set.
Each training text sample in the training text sample set corresponds to an emotion tendency label, and each training text sample comprises a category word segmentation. For example, comment text including category segmentation may be collected from a network, and related emotional tendencies, such as positive emotional tendencies, other emotional tendencies, negative emotional tendencies, etc., may be manually noted for specific objects in the comment text.
S220, training the emotion analysis model according to the training text sample set.
The emotion analysis model may be an initial emotion analysis model. In the embodiment of the application, an initial emotion analysis model can be trained for the category to which the object word belongs, for example, training is performed by utilizing the collected corpus comprising the category word.
Because the semantic style and sentence pattern similarity of the network text where the object word belongs to one class word are larger, in order to make the emotion analysis model more general and better utilize the existing data, the emotion analysis model in the embodiment of the application directly carries out more targeted emotion analysis on the class to which the object word belongs.
For the prior art, the training method using the neural network only trains through limited corpus, only aims at specific objects in general texts, and when the model faces specific objects which do not appear in the training set, the problem that the accuracy is low, the generalization capability of the model is not strong and the generalization effect is poor is easily caused because different analysis objects can cause different language styles and structures of the network texts.
According to the method, a traditional mode of forming sentence vectors by means of word vector splicing is not used on a neural network bottom layer structure, the generalization of a pre-training language model is utilized, the emotion analysis model is pre-trained by means of related evaluation papers outside a game, then the text and the category of a specific object are used as a reference unit for analysis at the emotion analysis model bottom layer, the characteristic that network new words are easy to appear in network text is considered through the training process of the emotion analysis model aiming at the category of the specific object, and the universality and the robustness of the emotion analysis model are enhanced.
The above steps are described in detail below.
In some embodiments, a sample segmented by a category corresponding to the object segmentation can be replaced as a training text sample set. As an example, the step S210 may include the steps of:
step k, according to object word segmentation corresponding to a first number of categories in the preset game field, filtering a second original corpus in the game field, and determining an initial text sample set; each initial text sample in the set of initial text samples includes an object segmentation.
And m, replacing the object word segmentation of each initial text sample in the initial text sample set with the category word segmentation corresponding to the object word segmentation to obtain an intermediate text sample set.
And n, determining a training text sample set according to the intermediate text sample set and the emotion tendency labels of each intermediate text sample.
In practical application, the game network comment data can be collected, the collected network comment data is subjected to relevant filtering, and only comment data containing specific objects is reserved. Since even the same specific object is in different types of games, the specific categories that may be represented are different. Therefore, before labeling the emotion tendency label, a specific object in the network comment data is replaced by a specific category corresponding to the specific object.
For example, only emotional tendency relationships between this web text and a particular category may be considered during the annotation process, where emotional tendency may be categorized as positive, other, and negative. For example, the set of training text samples for end use is shown in the following table:
in the embodiment of the application, the sample after the class segmentation corresponding to the object segmentation is replaced as the training text sample set, so that the model can be more aimed at the class corresponding to the object segmentation in the training process, but not the object segmentation itself, and the generalization and the universality of the emotion analysis model are enhanced.
In some embodiments, only comment data that contains a particular object in the training sample, or that can be clustered into a set of all object word segmentation clusters, may be retained. As one example, each of the first number of categories corresponds to a set of object segmentation clusters; the step k may include the steps of:
step o, word segmentation is carried out on each original text in the second original corpus;
and step p, deleting the original text which does not exist in the second original corpus and is contained in all the object word segmentation clustering sets and does not exist the word segmentation in all the object word segmentation clustering sets, thereby obtaining an initial text sample set.
In practical application, relevant filtering is carried out on the collected texts such as the network comment data, original texts which are not contained in all the object word segmentation cluster sets and are clustered to the words in all the object word segmentation cluster sets are deleted, and only the texts which contain the object words in the object word segmentation cluster sets or the original texts which can be clustered to the words in all the object word segmentation cluster sets are reserved.
The method has the advantages that the object word in the object word cluster set is not contained by filtering, the text in all the object word cluster sets cannot be clustered, the word in the obtained text sample set can be ensured to be classified into the category word corresponding to the object word cluster set, and the influence on the efficiency of the analysis process due to the fact that the category cannot be determined in the subsequent analysis process is avoided.
In some embodiments, as shown in FIG. 3, the final emotion analysis model may be obtained by training a language model and a classification model. As an example, the pre-trained emotion analysis model includes a pre-trained language model and a pre-trained classification model, and the step S220 may include the steps of:
And q, sequentially selecting a current training text sample from the training text sample set, performing the following steps r, s and t shown in fig. 3 until the analysis result of the emotion analysis model reaches the expectation, and outputting a pre-trained emotion analysis model.
And r, inputting the current training text sample and the current category word of the current training text sample into a pre-trained language model, and outputting a first sentence vector of the current training text sample and a second sentence vector of the current category word.
And step s, inputting the first sentence vector and the second sentence vector into a classification model to obtain the preliminary emotion tendencies.
And t, optimizing a pre-trained language model and a pre-trained classification model based on the current emotion tendency label corresponding to the initial emotion tendency and the current training sample by taking the difference between the minimized initial emotion tendency and the current emotion tendency label as a target, and taking the optimized language model and the optimized classification model as a new emotion analysis model to continue training.
In the step s, the preliminary emotion tendencies are emotion tendencies predicted by the classification model.
For the step t, it may be exemplified that whether the analysis result of the classification model reaches the expectation may be determined based on the current emotion tendency label corresponding to the initial emotion tendency and the current training sample; if the expected emotion tendency is not reached, optimizing the pre-trained language model and the optimized classification model based on the initial emotion tendency and the current emotion tendency label, and taking the optimized pre-trained language model and the optimized classification model as a new emotion analysis model to continue training; if the expectation is reached, the pre-trained language model and the classification model are directly used as the final pre-trained emotion analysis model.
In the embodiment of the application, training of the fine granularity emotion analysis model can be performed according to the training text sample set with the emotion tendency labels in the step S210. The training text sample may be an original web text including an object word, and the training text sample may further include a specific category corresponding to the object word, so as to perform model training with more category pertinence according to the training text sample including the category word. By utilizing the training text samples, the training process can be performed more systematically by optimizing the language model and the classification model, so that more accurate model parameters are obtained, and the accuracy of the parameters in the finally obtained emotion analysis model is higher.
It should be noted that the pre-trained language model may be an existing pre-trained language model, and in the embodiment of the present application, only the model may be fine-tuned. The following description will take a pre-trained language model as an example of a pre-trained Bert language model.
In practical application, the Bert language model has twelve layers, the dimension of the embedded vector is 768 dimensions, and the neural network dropout parameter is 0.1. The input of the Bert language model during training takes Chinese characters as units, and takes the shielding language model (Masked Language Model) as a training target to obtain the parameters of the related language model. In use, the output vector of the last layer of the pretrained Bert language model is used as the sentence vector represented by the input character.
Next, the web text is output as a first sentence vector of the web text at the last layer obtained by the pre-trained language model. For example, when the input text is "the fashion is fun", after the pre-trained language model is passed, the last layer of vectors of latitude (128, 768) of the model is used as the first sentence vector of the input text "the fashion is fun".
And outputting the last layer of the specific category obtained in the pre-trained language model as a second sentence vector of the specific category. The pre-trained language model can realize the word ambiguity problem which cannot be solved by the traditional word vector mode, and the corpus of the pre-trained language model is far greater than the training corpus of fine-granularity emotion analysis, so that better initialization can be brought to the subsequent classification model, and the sentence meaning of the text can be better and more completely represented.
And then, utilizing the first sentence vector and the second sentence vector obtained in the process, performing fine adjustment on the pre-trained language model according to the emotion tendency label based on the mutual weight influence between the two sentence vectors through the pre-trained classification model, and performing optimization training on the pre-trained classification model, wherein cross entropy is used for training loss. Training until the loss converges, and taking the trained language model and the trained classification model as a final fine granularity emotion analysis model.
In the embodiment of the application, the bottom layer of the model adopts words as objects to analyze sentences and specific objects as reference units, which is different from the existing analysis method which takes words as units, and the problem of error accumulation caused by word segmentation can be better avoided by taking words as units.
Furthermore, by training the pre-trained language model, a better initialization process can be provided for the subsequent classification model, so that sentence meaning of a text can be better and more complete, and by training the pre-trained classification model, the finally obtained emotion analysis model can be more in accordance with the accurate emotion tendency of the text.
In some embodiments, emotional tendency may be divided into three aspects. Based on this, emotion tendencies labels can include positive, other, and negative. The analysis result of the emotion tendencies of the text is more clear, and the true emotion tendencies can be conveniently and rapidly distinguished.
Fig. 4 provides a schematic structural diagram of an emotion analysis device in the game field. As shown in fig. 4, emotion analysis device 400 in the game field includes:
a first determining module 401, configured to determine, according to a first number of categories in a preset game field, a first object word segment belonging to a first category in word segments of a first text in the game field to be analyzed;
A replacing module 402, configured to replace a first object word in the first text with a category word of a first category, to obtain a second text;
and a prediction module 403, configured to apply a pre-trained emotion analysis model, and perform emotion prediction according to the second text, so as to obtain emotion tendencies corresponding to the first category.
In some embodiments, each of the first number of categories corresponds to a set of object segmentation clusters; the first determining module 401 is specifically configured to:
word segmentation is carried out on the first text;
matching the word segmentation in the first text in all object word segmentation clustering sets corresponding to the first number of categories;
if the first object segmentation in the first text is successfully matched in the first object segmentation cluster set, determining that the first object segmentation belongs to a first category corresponding to the first object segmentation cluster set.
In some embodiments, the apparatus further comprises:
the clustering module is used for clustering according to the word segmentation clustering set of all the objects and the words in the first text if the words in the first text are not successfully matched in the word segmentation clustering set of all the objects;
and the second determining module is used for determining that the first object segmentation word belongs to a first category corresponding to the first object segmentation word cluster set if the first object segmentation word in the first text is determined to be clustered into the first object segmentation word cluster set.
In some embodiments, the apparatus further comprises:
the preprocessing module is used for preprocessing a first original corpus in the game field based on word frequency and word part to obtain an object word segmentation set;
and the third determining module is used for clustering the object word segmentation set according to the first number of categories and determining the object word segmentation cluster set corresponding to each category in the first number of categories.
In some embodiments, the first number of categories includes: play, activity, occupation, important roles, props, player behavior, version, and teams.
In some embodiments, the pre-trained emotion analysis model includes a pre-trained language model and a pre-trained classification model; the prediction module 403 is specifically configured to:
converting the second text into sentence vectors by applying a pre-trained language model;
applying a pre-trained classification model, and determining emotion tendencies corresponding to the first category according to sentence vectors; among these emotional trends include positive, other, and negative.
The emotion analysis device in the game field provided by the embodiment of the application has the same technical characteristics as the emotion analysis method in the game field provided by the embodiment, so that the same technical problems can be solved, and the same technical effects can be achieved.
FIG. 5 provides a schematic structural diagram of an emotion analysis model training device in the field of games. As shown in fig. 5, emotion analysis model training device 500 in the game field includes:
a determining unit 501, configured to determine a training text sample set, where each training text sample in the training text sample set corresponds to an emotion tendency label, and each training text sample includes a category word;
the training unit 502 is configured to train the emotion analysis model according to the training text sample set.
In some embodiments, the determining unit 501 is specifically configured to:
filtering a second original corpus of the game field according to object segmentation corresponding to a first number of categories of the preset game field, and determining an initial text sample set, wherein each initial text sample in the initial text sample set comprises the object segmentation;
replacing the object word segmentation of each initial text sample in the initial text sample set with the category word segmentation corresponding to the object word segmentation to obtain an intermediate text sample set;
and determining a training text sample set according to the intermediate text sample set and the emotion tendency labels of each intermediate text sample.
In some embodiments, each of the first number of categories corresponds to a set of object segmentation clusters; the determining unit 501 is further configured to:
Word segmentation is carried out on each original text in the second original corpus;
deleting the original text which does not exist in the second original corpus and is contained in all the object word segmentation clustering sets and does not exist the word segments clustered into all the object word segmentation clustering sets, and obtaining an initial text sample set.
In some embodiments, training unit 502 is specifically configured to:
sequentially selecting a current training text sample from the training text sample set, and performing the following steps until the analysis result of the emotion analysis model reaches the expectation, and outputting a pre-trained emotion analysis model; the pre-trained emotion analysis model comprises a pre-trained language model and a pre-trained classification model;
inputting the current training text sample and the current category word segmentation of the current training text sample into a pre-trained language model, and outputting a first sentence vector of the current training text sample and a second sentence vector of the current category word segmentation;
inputting the first sentence vector and the second sentence vector into a classification model to obtain a preliminary emotion tendency;
based on the initial emotion tendency and the current emotion tendency label corresponding to the current training text sample, optimizing a pre-trained language model and a pre-trained classification model with the aim of minimizing the difference between the initial emotion tendency and the current emotion tendency label, and taking the optimized language model and the optimized classification model as a new emotion analysis model to continue training.
In some embodiments, emotional tendency labels include positive, other, and negative.
The emotion analysis model training device in the game field provided by the embodiment of the application has the same technical characteristics as the emotion analysis model training method in the game field provided by the embodiment, so that the same technical problems can be solved, and the same technical effects can be achieved.
As shown in fig. 6, a computer device 600 provided in an embodiment of the present application includes: the system comprises a processor 601, a memory 602 and a bus, wherein the memory 602 stores machine-readable instructions executable by the processor 601, when the computer device is running, the processor 601 communicates with the memory 602 through the bus, and the processor 601 executes the machine-readable instructions to perform the steps of the emotion analysis method in the game field or the emotion analysis model training method in the game field.
Specifically, the memory 602 and the processor 601 can be general-purpose memories and processors, and are not particularly limited herein, and when the processor 601 runs a computer program stored in the memory 602, the emotion analysis method in the game field or the emotion analysis model training method in the game field can be executed.
The embodiment of the application also provides a computer readable storage medium which stores machine executable instructions, wherein the computer executable instructions cause a processor to execute the emotion analysis method in the game field or the emotion analysis model training method in the game field when the computer executable instructions are called and executed by the processor.
The emotion analysis device in the game field or the emotion analysis model training device in the game field provided by the embodiment of the application can be specific hardware on equipment or software or firmware installed on the equipment and the like. The device provided by the embodiment of the present application has the same implementation principle and technical effects as those of the foregoing method embodiment, and for the sake of brevity, reference may be made to the corresponding content in the foregoing method embodiment where the device embodiment is not mentioned. It will be clear to those skilled in the art that, for convenience and brevity, the specific operation of the system, apparatus and unit described above may refer to the corresponding process in the above method embodiment, which is not described in detail herein.
In the embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other manners. The above-described apparatus embodiments are merely illustrative, for example, the division of the units is merely a logical function division, and there may be other manners of division in actual implementation, and for example, multiple units or components may be combined or integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be through some communication interface, device or unit indirect coupling or communication connection, which may be in electrical, mechanical or other form.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In addition, each functional unit in the embodiments provided in the present application may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the mobile control method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
It should be noted that: like reference numerals and letters in the following figures denote like items, and thus once an item is defined in one figure, no further definition or explanation of it is required in the following figures, and furthermore, the terms "first," "second," "third," etc. are used merely to distinguish one description from another and are not to be construed as indicating or implying relative importance.
Finally, it should be noted that: the above examples are only specific embodiments of the present application, and are not intended to limit the scope of the present application, but it should be understood by those skilled in the art that the present application is not limited thereto, and that the present application is described in detail with reference to the foregoing examples: any person skilled in the art may modify or easily conceive of the technical solution described in the foregoing embodiments, or perform equivalent substitution of some of the technical features, while remaining within the technical scope of the present disclosure; such modifications, changes or substitutions do not depart from the spirit of the corresponding technical solutions. Are intended to be encompassed within the scope of the present application. Therefore, the protection scope of the application is subject to the protection scope of the claims.
Claims (15)
1. An emotion analysis method in the field of games, comprising:
determining a first object word segmentation belonging to a first category in word segmentation of a first text of the game field to be analyzed according to a first number of categories of the preset game field;
replacing the first object word segmentation in the first text with the category word segmentation of the first category to obtain a second text;
Applying a pre-trained emotion analysis model, and carrying out emotion prediction according to the second text to obtain emotion tendencies corresponding to the first category;
the pre-trained emotion analysis model comprises a pre-trained language model and a pre-trained classification model; the training process of the pre-trained emotion analysis model comprises the following steps:
inputting a training text sample and a class word of the training text sample into the language model to obtain a first sentence vector of the training text sample and a second sentence vector of the class word; the training text sample corresponds to an emotion tendency label, and the emotion tendency label is marked manually;
inputting the first sentence vector and the second sentence vector into the classification model to obtain a preliminary emotion tendency;
and optimizing the language model and the classification model based on the preliminary emotion tendencies and the emotion tendencies labels to obtain a trained emotion analysis model.
2. The method of claim 1, wherein each of the first number of categories corresponds to a set of object segmentation clusters; according to a preset first number of categories of the game field, determining a first object word segmentation belonging to the first category in word segmentation of a first text of the game field to be analyzed, wherein the step comprises the following steps:
Word segmentation is carried out on the first text;
matching the word segments in the first text in all object word segment clustering sets corresponding to the first number of categories;
if the first object segmentation in the first text is successfully matched in the first object segmentation cluster set, determining that the first object segmentation belongs to a first category corresponding to the first object segmentation cluster set.
3. The method as recited in claim 2, further comprising:
if the word segmentation in the first text is not successfully matched in all the object word segmentation clustering sets;
clustering according to the word segmentation clustering set of all the objects and the word segmentation in the first text;
and if the first object segmentation word cluster in the first text is determined to be in the first object segmentation word cluster set, determining that the first object segmentation word belongs to a first category corresponding to the first object segmentation word cluster set.
4. The method according to claim 2, further comprising, before the step of determining a first object segmentation belonging to the first category among the segmentations of the first text of the game field to be analyzed according to a first number of categories of the preset game field:
Preprocessing a first original corpus in the game field based on word frequency and word parts to obtain an object word segmentation set;
clustering the object word segmentation set according to a first number of categories, and determining an object word segmentation cluster set corresponding to each category in the first number of categories.
5. The method of claim 1, wherein the first number of categories comprises: play, activity, occupation, important roles, props, player behavior, version, and teams.
6. The method of claim 1, wherein the pre-trained emotion analysis model comprises a pre-trained language model and a pre-trained classification model; applying a pre-trained emotion analysis model, and carrying out emotion prediction according to the second text to obtain emotion tendencies corresponding to the first category, wherein the method comprises the following steps:
converting the second text into sentence vectors by applying a pre-trained language model;
applying a pre-trained classification model, and determining emotion tendencies corresponding to the first category according to the sentence vectors; wherein the emotional tendency includes positive, other, and negative.
7. An emotion analysis model training method in the field of games, which is characterized by comprising the following steps:
Determining a training text sample set, wherein each training text sample in the training text sample set corresponds to an emotion tendency label, the emotion tendency labels are marked by manpower, and each training text sample comprises category segmentation;
training the emotion analysis model according to the training text sample set;
the emotion analysis model comprises a language model and a classification model;
training the emotion analysis model according to the training text sample set, including:
inputting the training text sample and the category word into the language model to obtain a first sentence vector of the training text sample and a second sentence vector of the category word;
inputting the first sentence vector and the second sentence vector into the classification model to obtain a preliminary emotion tendency;
and optimizing the language model and the classification model based on the preliminary emotion tendencies and the emotion tendencies labels to obtain a trained emotion analysis model.
8. The method of claim 7, wherein the step of determining a set of training text samples comprises:
filtering a second original corpus of the game field according to object segmentation corresponding to a first number of categories of the preset game field, and determining an initial text sample set, wherein each initial text sample in the initial text sample set comprises the object segmentation;
Replacing the object word of each initial text sample in the initial text sample set with the category word corresponding to the object word to obtain an intermediate text sample set;
and determining a training text sample set according to the intermediate text sample set and the emotion tendency labels of each intermediate text sample.
9. The method of claim 8, wherein each category of the first number of categories corresponds to a set of object segmentation clusters; the method comprises the steps of filtering a second original corpus of the game field according to object word segmentation corresponding to a first number of categories of the preset game field, and determining an initial text sample set, wherein the steps comprise:
word segmentation is carried out on each original text in the second original corpus;
deleting the original text which does not exist in the second original corpus and is contained in all the object word segmentation cluster sets and does not exist the word segments clustered into all the object word segmentation cluster sets, and obtaining an initial text sample set.
10. The method of claim 7, wherein training the emotion analysis model based on the training text sample set comprises:
Sequentially selecting a current training text sample from the training text sample set, executing a training step until the analysis result of the emotion analysis model reaches the expectation, and outputting a pre-trained emotion analysis model; the pre-trained emotion analysis model comprises a pre-trained language model and a pre-trained classification model;
the training step comprises the following steps:
inputting the current training text sample and the current class word of the current training text sample into the pre-trained language model, and outputting a first sentence vector of the current training text sample and a second sentence vector of the current class word;
inputting the first sentence vector and the second sentence vector into a classification model to obtain a preliminary emotion tendency;
and optimizing the pre-trained language model and the classification model based on the preliminary emotion tendency and the current emotion tendency label corresponding to the current training text sample by taking the difference between the preliminary emotion tendency and the current emotion tendency label as a target, and taking the optimized language model and classification model as a new emotion analysis model to continue training.
11. The method of claim 7, wherein the emotional tendency labels include positive, other, and negative.
12. An emotion analysis device in a game field, comprising:
the determining module is used for determining first object word segmentation belonging to a first category in word segmentation of a first text of the game field to be analyzed according to a first number of categories of the preset game field;
the replacing module is used for replacing the first object word segmentation in the first text with the category word segmentation of the first category to obtain a second text;
the prediction module is used for applying a pre-trained emotion analysis model, and performing emotion prediction according to the second text to obtain emotion tendencies corresponding to the first category;
the pre-trained emotion analysis model comprises a pre-trained language model and a pre-trained classification model; the training process of the pre-trained emotion analysis model comprises the following steps: inputting a training text sample and a class word of the training text sample into the language model to obtain a first sentence vector of the training text sample and a second sentence vector of the class word, wherein the training text sample corresponds to an emotion tendency label, and the emotion tendency label is marked by manpower; inputting the first sentence vector and the second sentence vector into the classification model to obtain a preliminary emotion tendency; and optimizing the language model and the classification model based on the preliminary emotion tendencies and the emotion tendencies labels to obtain a trained emotion analysis model.
13. An emotion analysis model training device in the field of games, comprising:
the system comprises a determining unit, a judging unit and a judging unit, wherein the determining unit is used for determining a training text sample set, each training text sample in the training text sample set corresponds to an emotion tendency label, the emotion tendency labels are marked by manpower, and each training text sample comprises category segmentation;
the training unit is used for training the emotion analysis model according to the training text sample set;
the emotion analysis model comprises a language model and a classification model;
the training unit is specifically used for:
inputting the training text sample and the category word into the language model to obtain a first sentence vector of the training text sample and a second sentence vector of the category word;
inputting the first sentence vector and the second sentence vector into the classification model to obtain a preliminary emotion tendency;
and optimizing the language model and the classification model based on the preliminary emotion tendencies and the emotion tendencies labels to obtain a trained emotion analysis model.
14. A computer device comprising a memory, a processor, the memory having stored therein a computer program executable on the processor, characterized in that the processor, when executing the computer program, implements the steps of the method of any of the preceding claims 1 to 11.
15. A computer readable storage medium storing machine executable instructions which, when invoked and executed by a processor, cause the processor to perform the method of any one of claims 1 to 11.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911322985.9A CN111125360B (en) | 2019-12-19 | 2019-12-19 | Emotion analysis method and device in game field and model training method and device thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911322985.9A CN111125360B (en) | 2019-12-19 | 2019-12-19 | Emotion analysis method and device in game field and model training method and device thereof |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111125360A CN111125360A (en) | 2020-05-08 |
| CN111125360B true CN111125360B (en) | 2023-10-20 |
Family
ID=70500547
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911322985.9A Active CN111125360B (en) | 2019-12-19 | 2019-12-19 | Emotion analysis method and device in game field and model training method and device thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111125360B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111783453B (en) * | 2020-07-01 | 2024-05-21 | 支付宝(杭州)信息技术有限公司 | Text emotion information processing method and device |
| CN112307757B (en) * | 2020-10-28 | 2023-07-28 | 中国平安人寿保险股份有限公司 | Emotion analysis method, device, equipment and storage medium based on auxiliary task |
| CN112667796B (en) * | 2021-01-05 | 2023-08-11 | 网易（杭州）网络有限公司 | Dialogue reply method and device, electronic equipment and readable storage medium |
| CN113094482B (en) * | 2021-03-29 | 2023-10-17 | 中国地质大学（北京） | Lightweight semantic intelligent service adaptation training evolution method and system |
| CN114757310B (en) * | 2022-06-16 | 2022-11-11 | 山东海量信息技术研究院 | Emotion recognition model and training method, device, equipment and readable storage medium thereof |
| CN115034318B (en) * | 2022-06-17 | 2024-05-17 | 中国平安人寿保险股份有限公司 | Method, device, equipment and medium for generating title discrimination model |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103971677A (en) * | 2013-02-01 | 2014-08-06 | 腾讯科技（深圳）有限公司 | Acoustic language model training method and device |
| CN109684634A (en) * | 2018-12-17 | 2019-04-26 | 北京百度网讯科技有限公司 | Sentiment analysis method, apparatus, equipment and storage medium |
| CN110032724A (en) * | 2018-12-19 | 2019-07-19 | 阿里巴巴集团控股有限公司 | The method and device that user is intended to for identification |
-
2019
- 2019-12-19 CN CN201911322985.9A patent/CN111125360B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103971677A (en) * | 2013-02-01 | 2014-08-06 | 腾讯科技（深圳）有限公司 | Acoustic language model training method and device |
| CN109684634A (en) * | 2018-12-17 | 2019-04-26 | 北京百度网讯科技有限公司 | Sentiment analysis method, apparatus, equipment and storage medium |
| CN110032724A (en) * | 2018-12-19 | 2019-07-19 | 阿里巴巴集团控股有限公司 | The method and device that user is intended to for identification |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111125360A (en) | 2020-05-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111125360B (en) | Emotion analysis method and device in game field and model training method and device thereof | |
| CN114610515B (en) | Multi-feature log anomaly detection method and system based on log full semantics | |
| CN110852087B (en) | Chinese error correction method and device, storage medium and electronic device | |
| CN110717339B (en) | Semantic representation model processing method and device, electronic equipment and storage medium | |
| CN110020422B (en) | Feature word determining method and device and server | |
| CN109657054B (en) | Abstract generation method, device, server and storage medium | |
| CN112270196B (en) | Entity relationship identification method and device and electronic equipment | |
| CN106407236B (en) | A kind of emotion tendency detection method towards comment data | |
| Chang et al. | Research on detection methods based on Doc2vec abnormal comments | |
| CN107688630B (en) | Semantic-based weakly supervised microbo multi-emotion dictionary expansion method | |
| CN113392209A (en) | Text clustering method based on artificial intelligence, related equipment and storage medium | |
| Olabenjo | Applying naive bayes classification to *** play apps categorization | |
| CN112860896A (en) | Corpus generalization method and man-machine conversation emotion analysis method for industrial field | |
| CN114997288A (en) | Design resource association method | |
| CN115757695A (en) | Log language model training method and system | |
| CN114065749A (en) | Text-oriented Guangdong language recognition model and training and recognition method of system | |
| CN112667819A (en) | Entity description reasoning knowledge base construction and reasoning evidence quantitative information acquisition method and device | |
| CN112562736A (en) | Voice data set quality evaluation method and device | |
| Tarnpradab et al. | Attention based neural architecture for rumor detection with author context awareness | |
| CN115269846A (en) | Text processing method and device, electronic equipment and storage medium | |
| CN114979620A (en) | Video bright spot segment detection method and device, electronic equipment and storage medium | |
| CN114676699A (en) | Entity emotion analysis method and device, computer equipment and storage medium | |
| CN114116967A (en) | Data cleaning method and device, electronic equipment and computer readable medium | |
| CN112632229A (en) | Text clustering method and device | |
| CN113420127A (en) | Threat information processing method, device, computing equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |