WO2020258481A1 - 个性化文本智能推荐方法、装置及计算机可读存储介质 - Google Patents
个性化文本智能推荐方法、装置及计算机可读存储介质 Download PDFInfo
- Publication number
- WO2020258481A1 WO2020258481A1 PCT/CN2019/102201 CN2019102201W WO2020258481A1 WO 2020258481 A1 WO2020258481 A1 WO 2020258481A1 CN 2019102201 W CN2019102201 W CN 2019102201W WO 2020258481 A1 WO2020258481 A1 WO 2020258481A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- words
- category
- corpus
- keyword
- word
- 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.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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
- G06F16/33—Querying
- G06F16/335—Filtering based on additional data, e.g. user or group profiles
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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
- G06F16/353—Clustering; Classification into predefined classes
Definitions
- This application relates to the field of artificial intelligence technology, and in particular to a personalized text recommendation method, device, and computer-readable storage medium.
- This application provides a personalized text intelligent recommendation method, device, and computer readable storage medium, the main purpose of which is to accurately recommend text data similar to the text data content when the user inputs text data.
- a personalized text intelligent recommendation method includes: receiving a corpus and a tag set including a basic text data set and a scene text data set, and performing word segmentation and deactivation on the corpus
- the preprocessing operation of words obtains a standard corpus; based on the keyword extraction algorithm, keyword extraction is performed on the standard corpus to obtain a keyword data set, cluster analysis is performed on the keyword data set to obtain a category set, and the Performing word vectorization operations on the category set and the keyword data set to obtain a word vector category set and a word vector keyword set; the word vector keyword set is input to a recommendation model, and the recommendation model uses the word vector keyword
- the set is input into the convolutional neural network for training and the first training value and the second training value are obtained.
- the first training value and the word vector category set are input into the loss function, and the loss function is calculated to obtain the first Loss value, judging the size of the first loss value and the preset first threshold value, if the first loss value is greater than the preset first threshold value, the convolutional neural network continues training, if the first loss value Is less than the preset first threshold, input the second training value and the label set into the loss function to calculate a second loss value, and determine the magnitude of the second loss value and the preset second threshold, if If the second loss value is greater than the preset second threshold, the convolutional neural network continues training, and if the second loss value is less than the preset second threshold, the convolutional neural network exits training; receiving text input by the user Data, inputting the text data into the recommendation model to determine the category and label of the text data, and recommending text content of the same category and label from the database according to the determined category and label.
- the present application also provides a personalized text intelligent recommendation device, which includes a memory and a processor, and the memory stores a personalized text intelligent recommendation program that can run on the processor,
- the personalized text intelligent recommendation program executes by the processor, the following steps are implemented: receiving a corpus and a tag set including a basic text data set and a scene text data set, and performing word segmentation and stop word removal on the corpus
- the standard corpus is obtained by the preprocessing operation;
- the keyword data set is obtained after keyword extraction is performed on the standard corpus based on the keyword extraction algorithm, and the cluster analysis is performed on the keyword data set to obtain the category set, and the category Performing word vectorization operations on the set and the keyword data set to obtain a word vector category set and a word vector keyword set;
- the word vector keyword set is input to a recommendation model, and the recommendation model uses the word vector keyword set Input to the convolutional neural network for training and obtain the first training value and the second training value, input the first training value and the word vector category set into the
- the convolutional neural network continues training, if the second loss value is less than the preset second threshold value, the convolutional neural network exits training; receiving text data input by the user , Inputting the text data into the recommendation model to determine the category and label of the text data, and recommending text content of the same category and label from the database according to the determined category and label.
- this application also provides a computer-readable storage medium with a personalized text intelligent recommendation program stored on the computer-readable storage medium, and the personalized text intelligent recommendation program can be used by one or more The processor executes to implement the steps of the personalized text intelligent recommendation method as described above.
- the personalized text intelligent recommendation method, device and computer-readable storage medium proposed in this application divides text data into basic text and scene text, which improves the precise division of text data content at the initial stage; at the same time, a probabilistic word segmentation model is constructed and the probability word segmentation model is maximized, which improves the feature extraction of the text data, which is efficient and maximum.
- the existing features can be utilized in a more advanced way; in addition, a convolutional neural network based on deep learning can effectively use the features for learning and improve the ability to recommend text data. Therefore, this application can realize accurate personalized text recommendation for users.
- FIG. 1 is a schematic flowchart of a personalized text intelligent recommendation method provided by an embodiment of this application
- FIG. 2 is a schematic diagram of the internal structure of a personalized text intelligent recommendation device provided by an embodiment of the application;
- FIG. 3 is a schematic diagram of modules of a personalized text smart recommendation program in a personalized text smart recommendation device provided by an embodiment of the application.
- This application provides a personalized text intelligent recommendation method.
- FIG. 1 it is a schematic flowchart of a personalized text intelligent recommendation method provided by an embodiment of this application.
- the method can be executed by a device, and the device can be implemented by software and/or hardware.
- the personalized text intelligent recommendation method includes:
- S1 Receive a corpus and a label set, and perform preprocessing operations including word segmentation and stop words removal on the corpus to obtain a standard corpus.
- the corpus in the preferred embodiment of the present application includes text data, and the corpus can be divided into a basic text data set and a scene text data set.
- the basic text data set includes a Weibo comment collection, a movie review collection, a music comment collection, and the like.
- the Weibo comment collection, the movie review collection, and the music comment collection all include several pieces of data.
- the scene text data set includes a stock comment collection, a government work report comment collection, a company quarterly annual financial statement comment collection, and a university student employment status comment collection.
- the tag set indicates the field of each text data in the corpus.
- text data "The Viking virus is mainly spread through network sharing. The virus infects all the .EXE executable files in the computer. The spread speed is very fast. After the Viking virus runs, modify the registry self-starting item to make itself follow the system Run together and generate the following virus files in the system file directory", which are marked as "computer domain” in the label set.
- the word segmentation includes establishing a probabilistic word segmentation model P(S) and maximizing the probability word segmentation model P(S) according to the corpus, and using the maximized probability word segmentation model P( S) Perform word segmentation operations on the corpus.
- the probability word segmentation model P(S) is:
- W 1 , W 2 ,..., W m are the words included in the corpus, m is the number of the corpus, and p(W i
- count(W i-1 ,W i ) represents the number of words W i-1 and W i appearing in the same text in the corpus at the same time
- count(W i-1 ) represents the word W i-1
- argmax represents the maximization operation.
- stop words are words that have no actual meaning in the text data and have no effect on the sentiment analysis of the text, but words that appear frequently.
- the stop words include commonly used pronouns, prepositions, and the like.
- the user A’s rating is concentrated: In fact, the supreme treasure in Journey to the West is like us in reality. We all thought that we would become the hero of the world and that we were the only one in this world, but gradually discovered that, I am no different from others. Suddenly looking back, it turned out that the one who missed the most was who we were. One day when you are walking, others will point to your back and say, "He is like a dog! So this movie is really interesting and thought-provoking.
- the method for removing stop words is a stop word list filtering method, which is based on the stop word list that has been constructed and the words in the corpus are matched one by one. If the matching is successful, the The word is a stop word, and the word is deleted from the corpus.
- the keyword extraction algorithm in the preferred embodiment includes: calculating the dependency correlation degree Dep(W i , W j ) between any two words W i , W j in the standard corpus:
- len(W i , W j ) represents the length of the dependency path between words W i and W j
- b is a hyperparameter
- tfidf (W i), tfidf (W j) denotes the word W i, W j of term frequency - inverse document frequency index, d represents the Euclidean distance between vectors of words W i and W j words of;
- a word with a large weight coefficient weight(W i , W j ) is selected according to the size of the weight coefficient, and the keyword extraction is completed to obtain a keyword data set.
- the cluster analysis described in the preferred embodiment of the present application includes randomized category center positions and optimized category center positions.
- the randomized category center position includes determining the number of category centers and randomly generating the coordinate positions of the category centers, and the number of category centers is the sum of the type of the basic text data set and the type of the scene text data set .
- the center position of the optimized category is:
- x i , x j are the data of the standard corpus
- dist(x i , x j ) is the positional distance between the data of the standard corpus
- D is the number of the center of the category.
- the category set is a text set with similar texts obtained after the cluster analysis.
- text data A in the standard corpus I'm really angry. I don't mind paying for it with feelings, but I don't want to pay for it without sincerity.
- the lines, performance, plot, and emotional changes of characters are not at the standard, and the transition editing is like fragmented pieces.
- the appearance gimmick turned out to be nothing more than a set of tidbits. I hope you will stop consuming yourself in the future. If you want to add a period of time, I hope that until the universe is destroyed.
- Text data B Rubbish, rubbish, rubbish, it will sell feelings, and now even feelings are not easy to sell, there is no content at all, only appearance gimmicks, placement of advertisements, editing is as broken as watching PPT, there is no plot and no performance.
- the lines are stiff to white. Since the text data A and the text data B are judged to have many identical words in the cluster analysis and belong to the same category, they are classified into the same category set.
- the word vectorization operation uses the Word2Vec algorithm.
- the Word2Vec algorithm includes an input layer, a projection layer, and an output layer.
- the input layer receives the keyword data set, and the output layer outputs the word vector.
- the projection layer ⁇ ( ⁇ ,j) is:
- the Huffman coding in the preferred embodiment of the present application uses different arrangements of 0 and 1 codes to represent the keyword data set according to data communication knowledge.
- the convolutional neural network includes a convolutional layer, a pooling layer, a first fully connected layer and a second fully connected layer.
- the convolution layer receives the word vector keyword set and performs a convolution operation on the word vector keyword set to obtain a convolution set.
- ⁇ ' is the convolution set
- ⁇ is the word vector keyword set
- k is the size of the convolution kernel
- s is the stride of the convolution operation
- p is the data zero-filling matrix
- the convolution set is input to the pooling layer, and the pooling layer searches for the word vector with the largest value of each word vector in the convolution set and forms a pooling set.
- the pooled set is simultaneously input to the first fully connected layer and the second fully connected layer, and the first fully connected layer and the second fully connected layer output the training value according to the activation function .
- the activation function is:
- y is the first training value or the second training value
- e is an infinite non-cyclic decimal
- the first loss value E1 is:
- ⁇ j is the word vector category set
- m is the number of the category set.
- the preset first threshold in the preferred embodiment of the present application is generally set to 0.5.
- the convolutional neural network classifies each keyword in the keyword data set into categories and the cluster analysis obtains the category set error If it is larger, it proves that the convolutional neural network has poor class recognition ability and needs to continue training.
- the second loss value E2 in the preferred embodiment of the present application is:
- x is the second training value
- ⁇ j is the label set
- m is the number of the label set.
- the preset second threshold is generally set to 0.01.
- the second loss value is greater than the preset second threshold, it indicates that the field classification of each keyword in the keyword data set by the convolutional neural network has a large error with the label set. If the convolutional neural network receives the "operating system” keyword, the "operating system” keyword is marked as the "computer” field in the label set, but the convolutional neural network may change the The keyword "operating system” is recognized as the "art” field, which indicates that the recognition ability of the convolutional neural network field is poor, and training is required.
- S8 Receive text data input by the user, input the text data into the recommendation model to determine the category and label of the text data, and recommend text content of the same category and label according to the determined category and label.
- the recommendation model obtains text data that has more of the same words as the text data X, and at the same time, analyzes the label of the text data X to input the NBA sports category, so it will recommend intelligently The text of the same category and the same label is output for users to read.
- the invention also provides a personalized text intelligent recommendation device.
- FIG. 2 it is a schematic diagram of the internal structure of a personalized text intelligent recommendation device provided by an embodiment of this application.
- the personalized text intelligent recommendation device 1 may be a PC (Personal Computer, personal computer), or a terminal device such as a smart phone, a tablet computer, or a portable computer, or a server.
- the personalized text intelligent recommendation device 1 at least includes a memory 11, a processor 12, a communication bus 13, and a network interface 14.
- the memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), magnetic memory, magnetic disk, optical disk, etc.
- the memory 11 may be an internal storage unit of the personalized text smart recommendation device 1 in some embodiments, for example, the hard disk of the personalized text smart recommendation device 1.
- the memory 11 may also be an external storage device of the personalized text smart recommendation device 1, such as a plug-in hard disk or a smart media card (SMC) equipped on the personalized text smart recommendation device 1. Secure Digital (SD) card, flash card (Flash Card), etc.
- SD Secure Digital
- flash card flash card
- the memory 11 may also include both an internal storage unit of the personalized text intelligent recommendation device 1 and an external storage device.
- the memory 11 can be used not only to store application software and various data installed in the personalized text smart recommendation device 1, such as the code of the personalized text smart recommendation program 01, but also to temporarily store data that has been output or will be output. .
- the processor 12 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip, and is used to run the program code or processing stored in the memory 11 Data, such as the execution of personalized text intelligent recommendation program 01, etc.
- CPU central processing unit
- controller microcontroller
- microprocessor or other data processing chip
- the communication bus 13 is used to realize the connection and communication between these components.
- the network interface 14 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface), and is usually used to establish a communication connection between the device 1 and other electronic devices.
- the device 1 may also include a user interface.
- the user interface may include a display (Display) and an input unit such as a keyboard (Keyboard).
- the optional user interface may also include a standard wired interface and a wireless interface.
- the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode, organic light emitting diode) touch device, etc.
- the display can also be appropriately called a display screen or a display unit, which is used to display the information processed in the personalized text intelligent recommendation device 1 and to display a visualized user interface.
- Figure 2 only shows the personalized text smart recommendation device 1 with components 11-14 and personalized text smart recommendation program 01. Those skilled in the art can understand that the structure shown in Figure 1 does not constitute a personalized text
- the definition of the smart recommendation device 1 may include fewer or more components than shown, or a combination of certain components, or a different component arrangement.
- a personalized text smart recommendation program 01 is stored in the memory 11; when the processor 12 executes the personalized text smart recommendation program 01 stored in the memory 11, the following steps are implemented:
- Step 1 Receive a corpus and a tag set, and perform a preprocessing operation including word segmentation and stop-word removal on the corpus to obtain a standard corpus.
- the corpus in the preferred embodiment of the present application includes text data, and the corpus can be divided into a basic text data set and a scene text data set.
- the basic text data set includes a Weibo comment collection, a movie review collection, a music comment collection, and the like.
- the Weibo comment collection, the movie review collection, and the music comment collection all include several pieces of data.
- the scene text data set includes a stock comment collection, a government work report comment collection, a company quarterly annual financial statement comment collection, and a university student employment status comment collection.
- the tag set indicates the field of each text data in the corpus.
- text data "The Viking virus is mainly spread through network sharing. The virus infects all the .EXE executable files in the computer. The spread speed is very fast. After the Viking virus runs, modify the registry self-starting item to make itself follow the system Run together and generate the following virus files in the system file directory", which are marked as "computer domain” in the label set.
- the word segmentation includes establishing a probabilistic word segmentation model P(S) and maximizing the probability word segmentation model P(S) according to the corpus, and using the maximized probability word segmentation model P( S) Perform word segmentation operations on the corpus.
- the probability word segmentation model P(S) is:
- W 1 , W 2 ,..., W m are the words included in the corpus, m is the number of the corpus, and p(W i
- count(W i-1 ,W i ) represents the number of words W i-1 and W i appearing in the same text in the corpus at the same time
- count(W i-1 ) represents the word W i-1
- argmax represents the maximization operation.
- stop words are words that have no actual meaning in the text data and have no effect on the sentiment analysis of the text, but words that appear frequently.
- the stop words include commonly used pronouns, prepositions, and the like.
- the user A’s rating is concentrated: In fact, the supreme treasure in Journey to the West is like us in reality. We all thought that we would become the hero of the world and that we were the only one in this world, but gradually discovered that, I am no different from others. Suddenly looking back, it turned out that the one who missed the most was who we were. One day when you are walking, others will point to your back and say, "He is like a dog! So this movie is really interesting and thought-provoking.
- the method for removing stop words is a stop word list filtering method, which is based on the stop word list that has been constructed and the words in the corpus are matched one by one. If the matching is successful, the The word is a stop word, and the word is deleted from the corpus.
- Step 2 Perform keyword extraction on the standard corpus based on a keyword extraction algorithm to obtain a keyword data set, perform a cluster analysis on the keyword data set to obtain a category set, and compare the category set with the keyword
- the data set is subjected to word vectorization operation to obtain the word vector category set and the word vector keyword set.
- the keyword extraction algorithm in the preferred embodiment includes: calculating the dependency correlation degree Dep(W i , W j ) between any two words W i , W j in the standard corpus:
- len(W i , W j ) represents the length of the dependency path between words W i and W j
- b is a hyperparameter
- tfidf (W i), tfidf (W j) denotes the word W i, W j of term frequency - inverse document frequency index, d represents the Euclidean distance between vectors of words W i and W j words of;
- the word with the largest weight coefficient weight(W i , W j ) is selected according to the size of the weight coefficient, and the keyword extraction is completed to obtain a keyword data set.
- the cluster analysis described in the preferred embodiment of the present application includes randomized category center positions and optimized category center positions.
- the randomized category center position includes determining the number of category centers and randomly generating the coordinate positions of the category centers, and the number of category centers is the sum of the type of the basic text data set and the type of the scene text data set .
- the center position of the optimized category is:
- x i , x j are the data of the standard corpus
- dist(x i , x j ) is the positional distance between the data of the standard corpus
- D is the number of the center of the category.
- the category set is a text set with similar texts obtained after the cluster analysis.
- text data A in the standard corpus I'm really angry. I don't mind paying for it with feelings, but I don't want to pay for it without sincerity.
- the lines, performance, plot, and emotional changes of characters are not at the standard, and the transition editing is like fragmented pieces.
- the appearance gimmick turned out to be nothing more than a set of tidbits. I hope you will stop consuming yourself in the future. If you want to add a period of time, I hope that until the universe is destroyed.
- Text data B Rubbish, rubbish, rubbish, it will sell feelings, and now even feelings are not easy to sell, there is no content at all, only appearance gimmicks, placement of advertisements, editing is as broken as watching PPT, there is no plot and no performance.
- the lines are stiff to white. Since the text data A and the text data B are judged to have many identical words in the cluster analysis and belong to the same category, they are classified into the same category set.
- the word vectorization operation uses the Word2Vec algorithm.
- the Word2Vec algorithm includes an input layer, a projection layer, and an output layer.
- the input layer receives the keyword data set, and the output layer outputs the word vector.
- the projection layer ⁇ ( ⁇ ,j) is:
- the Huffman coding in the preferred embodiment of the present application uses different arrangements of 0 and 1 codes to represent the keyword data set according to data communication knowledge.
- Step 3 Input the word vector keyword set into a recommendation model, and the recommendation model inputs the word vector keyword set into a convolutional neural network for training and obtains the first training value and the second training value, and The first training value and the word vector category set are input into the loss function, and the loss function is calculated to obtain a first loss value.
- the convolutional neural network includes a convolutional layer, a pooling layer, a first fully connected layer and a second fully connected layer.
- the convolution layer receives the word vector keyword set and performs a convolution operation on the word vector keyword set to obtain a convolution set.
- ⁇ ' is the convolution set
- ⁇ is the word vector keyword set
- k is the size of the convolution kernel
- s is the stride of the convolution operation
- p is the data zero-filling matrix
- the convolution set is input to the pooling layer, and the pooling layer searches for the word vector with the largest value of each word vector in the convolution set and forms a pooling set.
- the pooled set is simultaneously input to the first fully connected layer and the second fully connected layer, and the first fully connected layer and the second fully connected layer output the training value according to the activation function .
- the activation function is:
- y is the first training value or the second training value
- e is an infinite non-cyclic decimal
- the first loss value E1 is:
- ⁇ j is the word vector category set
- m is the number of the category set.
- Step 4 Determine the magnitude of the first loss value and the preset first threshold value.
- the preset first threshold in the preferred embodiment of the present application is generally set to 0.5.
- step 3 If the first loss value is greater than the preset first threshold, return to step 3, and the convolutional neural network continues training.
- the convolutional neural network classifies each keyword in the keyword data set into categories and the cluster analysis obtains the category set error If it is larger, it proves that the convolutional neural network has poor class recognition ability and needs to continue training.
- Step 5 If the first loss value is less than a preset first threshold, further input the second training value and the label set to the loss function to calculate a second loss value.
- the second loss value E2 in the preferred embodiment of the present application is:
- x is the second training value
- ⁇ j is the label set
- m is the number of the label set.
- Step 6 Determine the magnitude of the second loss value and the preset second threshold value.
- the preset second threshold is generally set to 0.01.
- step 3 If the second loss value is greater than the preset second threshold, return to step 3, and the convolutional neural network continues training.
- the second loss value is greater than the preset second threshold, it indicates that the field classification of each keyword in the keyword data set by the convolutional neural network has a large error with the label set. If the convolutional neural network receives the "operating system” keyword, the "operating system” keyword is marked as the "computer” field in the label set, but the convolutional neural network may change the The keyword "operating system” is recognized as the "art” field, which indicates that the recognition ability of the convolutional neural network field is poor, and training is required.
- Step 7 If the second loss value is less than a preset second threshold, the convolutional neural network exits training.
- Step 8 Receive text data input by the user, input the text data into the recommendation model to determine the category and label of the text data, and recommend text content of the same category and label based on the determined category and label.
- the recommendation model obtains text data that has more of the same words as the text data X, and at the same time, analyzes the label of the text data X to input the NBA sports category, so it will recommend intelligently The text of the same category and the same label is output for users to read.
- the personalized text intelligent recommendation program can also be divided into one or more modules, and the one or more modules are stored in the memory 11 and are executed by one or more processors (in this embodiment). For example, it is executed by the processor 12) to complete this application.
- the module referred to in this application refers to a series of computer program instruction segments that can complete specific functions, and is used to describe the personalized text intelligent recommendation program in the personalized text intelligent recommendation device The implementation process.
- the personalized text smart recommendation program can be divided into The source data receiving module 10, the feature extraction module 20, the feature analysis module 30, and the personalized text output module 40 are exemplary:
- the source data receiving module 10 is configured to receive a corpus and a tag set including a basic text data set and a scene text data set, and perform preprocessing operations including word segmentation and de-stop words on the corpus to obtain a standard corpus.
- the feature extraction module 20 is configured to: perform keyword extraction on the standard corpus based on a keyword extraction algorithm to obtain a keyword data set, perform cluster analysis on the keyword data set to obtain a category set, and divide the category Perform a word vectorization operation on the set and the keyword data set to obtain a word vector category set and a word vector keyword set.
- the feature analysis module 30 is configured to input the word vector keyword set into a recommendation model, and the recommendation model inputs the word vector keyword set into a convolutional neural network for training and obtains a first training value and
- the second training value, the first training value and the word vector category set are input into a loss function, the loss function is calculated to obtain a first loss value, and the difference between the first loss value and a preset first threshold is determined If the first loss value is greater than the preset first threshold, the convolutional neural network continues training, and if the first loss value is less than the preset first threshold, then the second training value is The label set is input to the loss function to calculate a second loss value, and the size of the second loss value and a preset second threshold is determined, and if the second loss value is greater than the preset second threshold, the convolution The neural network continues training, and if the second loss value is less than the preset second threshold, the convolutional neural network exits the training.
- the personalized text output module 40 is configured to: receive text data input by a user, input the text data into the recommendation model to determine the category and label of the text data, and determine the category and label of the text data according to the judged category and label. Recommend text content of the same category and label in the database.
- the embodiment of the present application also proposes a computer-readable storage medium, the computer-readable storage medium stores a personalized text intelligent recommendation program, and the personalized text intelligent recommendation program can be executed by one or more processors To achieve the following operations:
- a corpus and tag set including a basic text data set and a scene text data set, and performing preprocessing operations including word segmentation and removing stop words on the corpus to obtain a standard corpus;
- Keyword vectorization operation obtains word vector category set and word vector keyword set
- the word vector keyword set is input to a recommendation model, and the recommendation model inputs the word vector keyword set to a convolutional neural network for training and obtains the first training value and the second training value
- a training value and the word vector category set are input into a loss function, the loss function is calculated to obtain a first loss value, and the size of the first loss value and a preset first threshold is judged, if the first loss value Greater than the preset first threshold, the convolutional neural network continues training, and if the first loss value is less than the preset first threshold, the second training value and the label set are input to the loss function
- the second loss value is calculated, and the size of the second loss value and the preset second threshold is judged. If the second loss value is greater than the preset second threshold, the convolutional neural network continues training. If the first 2.
- the loss value is less than the preset second threshold, and the convolutional neural network exits training;
- the text data input by the user is received, the text data is input into the recommendation model to determine the category and label of the text data, and the text content of the same category and label is recommended from the database according to the determined category and label.
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)
- Computational Linguistics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
一种个性化文本智能推荐方法、装置以及计算机可读存储介质,涉及人工智能技术,实现精准的个性化文本推荐。所述方法包括:接收语料集和标签集,将所述语料集进行预处理得到标准语料集,对所述标准语料集进行关键字抽取后得到关键字数据集,对所述关键字数据集进行聚类分析得到类别集,将所述类别集与所述关键字数据集进行词向量化操作得到词向量类别集与词向量关键字集,将所述词向量类别集与所述词向量关键字集输入至推荐模型训练,直至所述推荐模型退出训练,接收用户输入的文本数据,判断所述文本数据的类别和标签,并根据判断的所述类别和标签从数据库中推荐相同类别和标签的文本内容。
Description
本申请要求于2019年6月28日提交中国专利局,申请号为201910582849.7、发明名称为“个性化文本智能推荐方法、装置及计算机可读存储介质”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
本申请涉及人工智能技术领域,尤其涉及一种个性化文本推荐方法、装置及计算机可读存储介质。
随着互联网信息的快速增长,用户每天会浏览大量的文本数据,如果能从用户所浏览的文本数据中提取有用的关键字信息,从而进行个性化推荐,则能更高效的利用计算资源,并节约用户时间。目前国内外学者对推荐算法进行了相关研究,研究发现,其中数据稀疏性问题、冷启动问题以及用户兴趣获取问题都是影响推荐效果的重要因素。因此基于所述研究结果,现有的一些网站例如:电影、音乐、小说等使用神经规则引擎方法进行个性化推荐,所述神经规则引擎方法虽然精确,但是其方法僵硬脆弱,推荐的内容往往与用户实际所需的内容大相径庭,因此,个性化推荐的准确率有待进一步加强。
发明内容
本申请提供一种个性化文本智能推荐方法、装置及计算机可读存储介质,其主要目的在于当用户输入文本数据时,给用户精准的推荐与所述文本数据内容相近的文本数据。
为实现上述目的,本申请提供的一种个性化文本智能推荐方法,包括:接收包括基础文本数据集和场景文本数据集的语料集和标签集,将所述语料集进行包括分词、去停用词的预处理操作得到标准语料集;基于关键字抽取算法对所述标准语料集进行关键字抽取后得到关键字数据集,对所述关键字数据集进行聚类分析得到类别集,将所述类别集与所述关键字数据集进行词 向量化操作得到词向量类别集与词向量关键字集;将所述词向量关键字集输入至推荐模型,所述推荐模型将所述词向量关键字集输入至卷积神经网络中进行训练并得到第一训练值和第二训练值,将所述第一训练值和所述词向量类别集输入至损失函数中,所述损失函数计算得到第一损失值,判断所述第一损失值与预设第一阈值的大小,若所述第一损失值大于预设第一阈值,则所述卷积神经网络继续训练,若所述第一损失值小于预设第一阈值,则将所述第二训练值和所述标签集输入至所述损失函数计算得到第二损失值,判断所述第二损失值与预设第二阈值的大小,若所述第二损失值大于预设第二阈值,所述卷积神经网络继续训练,若所述第二损失值小于预设第二阈值,所述卷积神经网络退出训练;接收用户输入的文本数据,将所述文本数据输入至所述推荐模型中判断所述文本数据的类别和标签,并根据判断的所述类别和标签从数据库中推荐相同类别和标签的文本内容。
此外,为实现上述目的,本申请还提供一种个性化文本智能推荐装置,该装置包括存储器和处理器,所述存储器中存储有可在所述处理器上运行的个性化文本智能推荐程序,所述个性化文本智能推荐程序被所述处理器执行时实现如下步骤:接收包括基础文本数据集和场景文本数据集的语料集和标签集,将所述语料集进行包括分词、去停用词的预处理操作得到标准语料集;基于关键字抽取算法对所述标准语料集进行关键字抽取后得到关键字数据集,对所述关键字数据集进行聚类分析得到类别集,将所述类别集与所述关键字数据集进行词向量化操作得到词向量类别集与词向量关键字集;将所述词向量关键字集输入至推荐模型,所述推荐模型将所述词向量关键字集输入至卷积神经网络中进行训练并得到第一训练值和第二训练值,将所述第一训练值和所述词向量类别集输入至损失函数中,所述损失函数计算得到第一损失值,判断所述第一损失值与预设第一阈值的大小,若所述第一损失值大于预设第一阈值,则所述卷积神经网络继续训练,若所述第一损失值小于预设第一阈值,则将所述第二训练值和所述标签集输入至所述损失函数计算得到第二损失值,判断所述第二损失值与预设第二阈值的大小,若所述第二损失值大于预设第二阈值,所述卷积神经网络继续训练,若所述第二损失值小于预设第二阈值,所述卷积神经网络退出训练;接收用户输入的文本数据,将所述文本数据输入至所述推荐模型中判断所述文本数据的类别和标签,并根据判断 的所述类别和标签从数据库中推荐相同类别和标签的文本内容。
此外,为实现上述目的,本申请还提供一种计算机可读存储介质,所述计算机可读存储介质上存储有个性化文本智能推荐程序,所述个性化文本智能推荐程序可被一个或者多个处理器执行,以实现如上所述的个性化文本智能推荐方法的步骤。
本申请提出的个性化文本智能推荐方法、装置及计算机可读存储介质。本申请将文本数据分为基础文本和场景文本,提高了初期对文本数据内容的精确划分;同时构建了概率分词模型并最大化概率分词模型,提高对所述文本数据的特征提取,高效并最大化的利用到已有特征;另外基于深度学习的卷积神经网络可有效的利用所述特征进行学习,提高对文本数据的推荐能力。因此,本申请可为用户实现精确的个性化文本推荐。
图1为本申请一实施例提供的个性化文本智能推荐方法的流程示意图;
图2为本申请一实施例提供的个性化文本智能推荐装置的内部结构示意图;
图3为本申请一实施例提供的个性化文本智能推荐装置中个性化文本智能推荐程序的模块示意图。
本申请目的的实现、功能特点及优点将结合实施例,参照附图做进一步说明。
应当理解,此处所描述的具体实施例仅仅用以解释本申请,并不用于限定本申请。
本申请提供一种个性化文本智能推荐方法。参照图1所示,为本申请一实施例提供的个性化文本智能推荐方法的流程示意图。该方法可以由一个装置执行,该装置可以由软件和/或硬件实现。
在本实施例中,个性化文本智能推荐方法包括:
S1、接收语料集和标签集,将所述语料集进行包括分词、去停用词的预处理操作得到标准语料集。
本申请较佳实施例所述语料集包括文本数据,所述语料集可分为基础文本数据集和场景文本数据集。
进一步地所述基础文本数据集包括微博评论集、影电观后感集、音乐评论集等。所述微博评论集、所述影电观后感集、所述乐评论集都包括若干条数据。所述场景文本数据集包括股票评论集、政府工作报告评论集、公司季度年度财务报表评论集、大学生就业情况评论集。
优选地,所述标签集注明所述语料集内各文本数据所属领域。如文本数据:“威金病毒主要通过网络共享传播,病毒会感染电脑中所有的.EXE可执行文件,传播速度十分迅速。威金病毒运行后,修改注册表自启动项,以使自己随系统一起运行,向系统文件目录下生成以下病毒文件”,在标签集中注明为“计算机领域”。
本申请较佳实施例中,所述分词包括根据所述语料集建立概率分词模型P(S)和最大化所述概率分词模型P(S),并利用最大化的所述概率分词模型P(S)对所述语料集执行分词操作。
其中,所述概率分词模型P(S)为:
其中,W
1,W
2,…,W
m为所述语料集包括的词,m为所述语料集的数量,p(W
i|W
i-1)表示在词W
i-1出现的情况下词W
i出现的概率;
所述最大化的所述概率分词模型P(S):
其中,count(W
i-1,W
i)表示词W
i-1和词W
i同时出现在所述语料集内同一篇文本的文本数量,count(W
i-1)表示词W
i-1出现在所述语料集内的文本数量,argmax表示最大化操作。
进一步所述停用词是文本数据中没有什么实际意义的词,且对文本的情感分析没有什么影响,但出现频率高的词,所述停用词包括常用的代词、介词等。如所述影电观后感集中用户A的影评为:其实大话西游中的至尊宝就像现实中的我们,都曾以为自己会成为盖世英雄,以为自己是这个世界的唯一,可是渐渐发现,自己与别人并没有什么不同。蓦然回首,原来最怀念的,是当初的我们。有一天你走着,别人会指着你的背影说:“他好像条狗啊!”,所以这部电影真的很耐人寻味发人深省。在去除停用词后则变为“其实大话 西游至尊宝像现实我们,都曾以为自己成为盖世英雄,以为自己这个世界唯一,渐渐发现,自己别人没有什么不同。蓦然回首,原来最怀念,是当初我们。有一天你走着,别人指着你背影说:“他好像条狗!”,所以这部电影真耐人寻味发人深省”
本申请较佳实施例,所述去停用词的方法为停用词表过滤法,基于已构建好的停用词表和所述语料集的词进行一一匹配,若匹配成功,则该词为停用词,且将所述该词从所述语料集中删除。
S2、基于关键字抽取算法对所述标准语料集进行关键字抽取后得到关键字数据集,对所述关键字数据集进行聚类分析得到类别集,将所述类别集与所述关键字数据集进行词向量化操作得到词向量类别集与词向量关键字集。
较佳实施例所述关键字抽取算法包括:计算所述标准语料集中任意两词W
i,W
j之间的依存关联度Dep(W
i,W
j):
其中,len(W
i,W
j)表示词语W
i和W
j之间的依存路径长度,b是超参数;
计算所述标准语料集中任意两词W
i,W
j之间的引力值f
grav(W
i,W
j):
其中,tfidf(W
i)、tfidf(W
j)表示词W
i,W
j的词频-逆文本频率指数,d表示词W
i和W
j的词向量之间的欧式距离;
根据所述依存关联度Dep(W
i,W
j)和所述引力值f
grav(W
i,W
j)判断所述标准语料集中任意两词W
i,W
j之间的权重系数weight(W
i,W
j):
weight(W
i,W
j)=Dep(W
i,W
j)*f
grav(W
i,W
j)
按照所述权重系数大小选择权重系数weight(W
i,W
j)大的词,完成所述关键字抽取,得到关键字数据集。
本申请较佳实施例所述聚类分析包括随机化类别中心位置和最优化类别中心位置。
其中,所述随机化类别中心位置包括确定类别中心个数和随机生成所述类别中心的坐标位置,所述类别中心个数为所述基础文本数据集的种类和场景文本数据集的种类的总和。
所述最优化类别中心位置为:
其中,x
i,x
j为所述标准语料集的数据,dist(x
i,x
j)为所述标准语料集数据之间的位置距离,D为所述类别中心个数。
较优地所述类别集是通过所述聚类分析后,得到的具有相似文本的文本集。如所述标准语料集中有文本数据A:真生气,我不介意情怀买单,但我不想毫无诚意情怀买单,台词、表演、剧情、人物情感变化拿捏都不在水准,转场剪辑像支离破碎拼凑。出场噱头原来不过片场花絮,到处植入硬广摆明圈钱标语。希望你以后停止消费自己,如果要加一段期限,我希望直到宇宙毁灭。文本数据B:垃圾,垃圾,垃圾,就会卖情怀,现在连情怀都不好好卖,一点内容都没有,只有出场噱头、植入广告,剪辑像是看PPT一样破碎不堪,没有剧情没有表演,台词对白僵硬。由于所述文本数据A与所述文本数据B在所述聚类分析中被判别有很多相同用词,属于相同类别,因此被划分为同一类别集中。
较佳实施例所述词向量化操作采用Word2Vec算法,所述Word2Vec算法包括输入层、投影层和输出层,所述输入层接收所述关键字数据集,所述输出层输出得到所述词向量集,所述投影层ζ(ω,j)为:
本申请较佳实施例所述霍夫曼编码是根据数据通信知识使用0,1码的不同排列来表示所述关键字数据集。
S3、将所述词向量关键字集输入至推荐模型,所述推荐模型将所述词向量关键字集输入至卷积神经网络中进行训练并得到第一训练值和第二训练值,将所述第一训练值和所述词向量类别集输入至所述损失函数中,所述损失函数计算得到第一损失值。
优选地所述卷积神经网络包括卷积层、池化层、第一全连接层和第二全连接层。所述卷积层接收所述词向量关键字集并对所述词向量关键字集进行卷积操作得到卷积集。
进一步地所述卷积操作为:
其中ω’为所述卷积集,ω为所述词向量关键字集,k为卷积核的大小,s为所述卷积操作的步幅,p为数据补零矩阵。
本申请较佳实施例将所述卷积集输入至所述池化层,所述池化层寻找所述卷积集中各词向量数值最大的词向量并组成池化集。
进一步地将所述池化集同时输入至所述第一全连接层和所述第二全连接层,所述第一全连接层和所述第二全连接层根据激活函数输出所述训练值。所述激活函数为:
其中y为所述第一训练值或第二训练值,e为无限不循环小数。
较佳地所述第一损失值E1为:
其中,x为所述第一训练值,μ
j为所述词向量类别集,m为所述类别集的数量。
S4、判断所述第一损失值与预设第一阈值的大小。
本申请较佳实施例所述预设第一阈值一般设定为0.5。
若所述第一损失值大于预设第一阈值,则返回S3,所述卷积神经网络继续训练。
当所述第一损失值大于所述预设第一阈值时,表明所述卷积神经网络对所述关键字数据集内各关键字的类别分类与所述聚类分析得到所述类别集误差较大,证明所述卷积神经网络识别类别能力较差,需继续训练。
S5、若所述第一损失值小于预设第一阈值,则进一步将所述第二训练值和所述标签集输入至所述损失函数计算得到第二损失值。
本申请较佳实施例所述第二损失值E2为:
其中,x为所述第二训练值,μ
j为所述标签集,m为所述标签集的数量。
S6、判断所述第二损失值与预设第二阈值的大小。
较佳地所述预设第二阈值一般设置为0.01。
若所述第二损失值大于预设第二阈值,返回S3,所述卷积神经网络继续训练。
若所述第二损失值大于预设第二阈值,表明所述卷积神经网络对所述关键字数据集内各关键字的领域分类与所述标签集误差较大。如所述卷积神经网络接受到“操作系统”关键字,所述“操作系统”关键字在所述标签集中注明为“计算机”领域,但所述卷积神经网络可能会将所述“操作系统”关键字识别为“艺术”领域,表明所述卷积神经网络领域识别能力较差,需继续训练。
S7、若所述第二损失值小于预设第二阈值,所述卷积神经网络退出训练。
S8、接收用户输入的文本数据,将所述文本数据输入至所述推荐模型中判断所述文本数据的类别和标签,并根据所述判断的类别和标签推荐相同类别和标签的文本内容。
较佳地,如用户输入文本数据X为:历经四年,终于要划上句号了,今年的骑士,堪称是史上最烂的总决赛参赛球队,能拼到今天已经是难能可贵,老套的说一句吧,是“虽败犹荣”。至于NBA的两大命题,东部詹姆斯厉害和联盟抗勇无敌则仍然是未解之谜,只能在下个赛季拭目以待了。所述推荐模型根据所述聚类分析后得出与所述文本数据X有较多相同用词的文本数据,同时分析出所述文本数据X的标签输入NBA体育类,因此会智能化的推荐出相同类别和相同标签的文本供用户阅读。
发明还提供一种个性化文本智能推荐装置。参照图2所示,为本申请一实施例提供的个性化文本智能推荐装置的内部结构示意图。
在本实施例中,所述个性化文本智能推荐装置1可以是PC(Personal Computer,个人电脑),或者是智能手机、平板电脑、便携计算机等终端设备,也可以是一种服务器等。该个性化文本智能推荐装置1至少包括存储器11、处理器12,通信总线13,以及网络接口14。
其中,存储器11至少包括一种类型的可读存储介质,所述可读存储介质包括闪存、硬盘、多媒体卡、卡型存储器(例如,SD或DX存储器等)、磁性存储器、磁盘、光盘等。存储器11在一些实施例中可以是个性化文本智能推荐装置1的内部存储单元,例如该个性化文本智能推荐装置1的硬盘。存储器11在另一些实施例中也可以是个性化文本智能推荐装置1的外部存储设 备,例如个性化文本智能推荐装置1上配备的插接式硬盘,智能存储卡(Smart Media Card,SMC),安全数字(Secure Digital,SD)卡,闪存卡(Flash Card)等。进一步地,存储器11还可以既包括个性化文本智能推荐装置1的内部存储单元也包括外部存储设备。存储器11不仅可以用于存储安装于个性化文本智能推荐装置1的应用软件及各类数据,例如个性化文本智能推荐程序01的代码等,还可以用于暂时地存储已经输出或者将要输出的数据。
处理器12在一些实施例中可以是一中央处理器(Central Processing Unit,CPU)、控制器、微控制器、微处理器或其他数据处理芯片,用于运行存储器11中存储的程序代码或处理数据,例如执行个性化文本智能推荐程序01等。
通信总线13用于实现这些组件之间的连接通信。
网络接口14可选的可以包括标准的有线接口、无线接口(如WI-FI接口),通常用于在该装置1与其他电子设备之间建立通信连接。
可选地,该装置1还可以包括用户接口,用户接口可以包括显示器(Display)、输入单元比如键盘(Keyboard),可选的用户接口还可以包括标准的有线接口、无线接口。可选地,在一些实施例中,显示器可以是LED显示器、液晶显示器、触控式液晶显示器以及OLED(Organic Light-Emitting Diode,有机发光二极管)触摸器等。其中,显示器也可以适当的称为显示屏或显示单元,用于显示在个性化文本智能推荐装置1中处理的信息以及用于显示可视化的用户界面。
图2仅示出了具有组件11-14以及个性化文本智能推荐程序01的个性化文本智能推荐装置1,本领域技术人员可以理解的是,图1示出的结构并不构成对个性化文本智能推荐装置1的限定,可以包括比图示更少或者更多的部件,或者组合某些部件,或者不同的部件布置。
在图2所示的装置1实施例中,存储器11中存储有个性化文本智能推荐程序01;处理器12执行存储器11中存储的个性化文本智能推荐程序01时实现如下步骤:
步骤一、接收语料集和标签集,将所述语料集进行包括分词、去停用词的预处理操作得到标准语料集。
本申请较佳实施例所述语料集包括文本数据,所述语料集可分为基础文本数据集和场景文本数据集。
进一步地所述基础文本数据集包括微博评论集、影电观后感集、音乐评论集等。所述微博评论集、所述影电观后感集、所述乐评论集都包括若干条数据。所述场景文本数据集包括股票评论集、政府工作报告评论集、公司季度年度财务报表评论集、大学生就业情况评论集。
优选地,所述标签集注明所述语料集内各文本数据所属领域。如文本数据:“威金病毒主要通过网络共享传播,病毒会感染电脑中所有的.EXE可执行文件,传播速度十分迅速。威金病毒运行后,修改注册表自启动项,以使自己随系统一起运行,向系统文件目录下生成以下病毒文件”,在标签集中注明为“计算机领域”。
本申请较佳实施例中,所述分词包括根据所述语料集建立概率分词模型P(S)和最大化所述概率分词模型P(S),并利用最大化的所述概率分词模型P(S)对所述语料集执行分词操作。
其中,所述概率分词模型P(S)为:
其中,W
1,W
2,…,W
m为所述语料集包括的词,m为所述语料集的数量,p(W
i|W
i-1)表示在词W
i-1出现的情况下词W
i出现的概率;
所述最大化的所述概率分词模型P(S):
其中,count(W
i-1,W
i)表示词W
i-1和词W
i同时出现在所述语料集内同一篇文本的文本数量,count(W
i-1)表示词W
i-1出现在所述语料集内的文本数量,argmax表示最大化操作。
进一步所述停用词是文本数据中没有什么实际意义的词,且对文本的情感分析没有什么影响,但出现频率高的词,所述停用词包括常用的代词、介词等。如所述影电观后感集中用户A的影评为:其实大话西游中的至尊宝就像现实中的我们,都曾以为自己会成为盖世英雄,以为自己是这个世界的唯一,可是渐渐发现,自己与别人并没有什么不同。蓦然回首,原来最怀念的,是当初的我们。有一天你走着,别人会指着你的背影说:“他好像条狗啊!”,所以这部电影真的很耐人寻味发人深省。在去除停用词后则变为“其实大话西游至尊宝像现实我们,都曾以为自己成为盖世英雄,以为自己这个世界唯一,渐渐发现,自己别人没有什么不同。蓦然回首,原来最怀念,是当初我 们。有一天你走着,别人指着你背影说:“他好像条狗!”,所以这部电影真耐人寻味发人深省”
本申请较佳实施例,所述去停用词的方法为停用词表过滤法,基于已构建好的停用词表和所述语料集的词进行一一匹配,若匹配成功,则该词为停用词,且将所述该词从所述语料集中删除。
步骤二、基于关键字抽取算法对所述标准语料集进行关键字抽取后得到关键字数据集,对所述关键字数据集进行聚类分析得到类别集,将所述类别集与所述关键字数据集进行词向量化操作得到词向量类别集与词向量关键字集。
较佳实施例所述关键字抽取算法包括:计算所述标准语料集中任意两词W
i,W
j之间的依存关联度Dep(W
i,W
j):
其中,len(W
i,W
j)表示词语W
i和W
j之间的依存路径长度,b是超参数;
计算所述标准语料集中任意两词W
i,W
j之间的引力值f
grav(W
i,W
j):
其中,tfidf(W
i)、tfidf(W
j)表示词W
i,W
j的词频-逆文本频率指数,d表示词W
i和W
j的词向量之间的欧式距离;
根据所述依存关联度Dep(W
i,W
j)和所述引力值f
grav(W
i,W
j)判断所述标准语料集中任意两词W
i,W
j之间的权重系数weight(W
i,W
j):
weight(W
i,W
j)=Dep(W
i,W
j)*f
grav(W
i,W
j)
按照所述权重系数大小选择权重系数weight(W
i,W
j)最大的词,完成所述关键字抽取,得到关键字数据集。
本申请较佳实施例所述聚类分析包括随机化类别中心位置和最优化类别中心位置。
其中,所述随机化类别中心位置包括确定类别中心个数和随机生成所述类别中心的坐标位置,所述类别中心个数为所述基础文本数据集的种类和场景文本数据集的种类的总和。
所述最优化类别中心位置为:
其中,x
i,x
j为所述标准语料集的数据,dist(x
i,x
j)为所述标准语料集数据之间的位置距离,D为所述类别中心个数。
较优地所述类别集是通过所述聚类分析后,得到的具有相似文本的文本集。如所述标准语料集中有文本数据A:真生气,我不介意情怀买单,但我不想毫无诚意情怀买单,台词、表演、剧情、人物情感变化拿捏都不在水准,转场剪辑像支离破碎拼凑。出场噱头原来不过片场花絮,到处植入硬广摆明圈钱标语。希望你以后停止消费自己,如果要加一段期限,我希望直到宇宙毁灭。文本数据B:垃圾,垃圾,垃圾,就会卖情怀,现在连情怀都不好好卖,一点内容都没有,只有出场噱头、植入广告,剪辑像是看PPT一样破碎不堪,没有剧情没有表演,台词对白僵硬。由于所述文本数据A与所述文本数据B在所述聚类分析中被判别有很多相同用词,属于相同类别,因此被划分为同一类别集中。
较佳实施例所述词向量化操作采用Word2Vec算法,所述Word2Vec算法包括输入层、投影层和输出层,所述输入层接收所述关键字数据集,所述输出层输出得到所述词向量集,所述投影层ζ(ω,j)为:
本申请较佳实施例所述霍夫曼编码是根据数据通信知识使用0,1码的不同排列来表示所述关键字数据集。
步骤三、将所述词向量关键字集输入至推荐模型,所述推荐模型将所述词向量关键字集输入至卷积神经网络中进行训练并得到第一训练值和第二训练值,将所述第一训练值和所述词向量类别集输入至所述损失函数中,所述损失函数计算得到第一损失值。
优选地所述卷积神经网络包括卷积层、池化层、第一全连接层和第二全连接层。所述卷积层接收所述词向量关键字集并对所述词向量关键字集进行卷积操作得到卷积集。
进一步地所述卷积操作为:
其中ω’为所述卷积集,ω为所述词向量关键字集,k为卷积核的大小,s为所述卷积操作的步幅,p为数据补零矩阵。
本申请较佳实施例将所述卷积集输入至所述池化层,所述池化层寻找所述卷积集中各词向量数值最大的词向量并组成池化集。
进一步地将所述池化集同时输入至所述第一全连接层和所述第二全连接层,所述第一全连接层和所述第二全连接层根据激活函数输出所述训练值。所述激活函数为:
其中y为所述第一训练值或第二训练值,e为无限不循环小数。
较佳地所述第一损失值E1为:
其中,x为所述第一训练值,μ
j为所述词向量类别集,m为所述类别集的数量。
步骤四、判断所述第一损失值与预设第一阈值的大小。
本申请较佳实施例所述预设第一阈值一般设定为0.5。
若所述第一损失值大于预设第一阈值,则返回步骤三,所述卷积神经网络继续训练。
当所述第一损失值大于所述预设第一阈值时,表明所述卷积神经网络对所述关键字数据集内各关键字的类别分类与所述聚类分析得到所述类别集误差较大,证明所述卷积神经网络识别类别能力较差,需继续训练。
步骤五、若所述第一损失值小于预设第一阈值,则进一步将所述第二训练值和所述标签集输入至所述损失函数计算得到第二损失值。
本申请较佳实施例所述第二损失值E2为:
其中,x为所述第二训练值,μ
j为所述标签集,m为所述标签集的数量。
步骤六、判断所述第二损失值与预设第二阈值的大小。
较佳地所述预设第二阈值一般设置为0.01。
若所述第二损失值大于预设第二阈值,返回步骤三,所述卷积神经网络继续训练。
若所述第二损失值大于预设第二阈值,表明所述卷积神经网络对所述关键字数据集内各关键字的领域分类与所述标签集误差较大。如所述卷积神经网络接受到“操作系统”关键字,所述“操作系统”关键字在所述标签集中注明为“计算机”领域,但所述卷积神经网络可能会将所述“操作系统”关键字识别为“艺术”领域,表明所述卷积神经网络领域识别能力较差,需继续训练。
步骤七、若所述第二损失值小于预设第二阈值,所述卷积神经网络退出训练。
步骤八、接收用户输入的文本数据,将所述文本数据输入至所述推荐模型中判断所述文本数据的类别和标签,并根据所述判断的类别和标签推荐相同类别和标签的文本内容。
较佳地,如用户输入文本数据X为:历经四年,终于要划上句号了,今年的骑士,堪称是史上最烂的总决赛参赛球队,能拼到今天已经是难能可贵,老套的说一句吧,是“虽败犹荣”。至于NBA的两大命题,东部詹姆斯厉害和联盟抗勇无敌则仍然是未解之谜,只能在下个赛季拭目以待了。所述推荐模型根据所述聚类分析后得出与所述文本数据X有较多相同用词的文本数据,同时分析出所述文本数据X的标签输入NBA体育类,因此会智能化的推荐出相同类别和相同标签的文本供用户阅读。
可选地,在其他实施例中,个性化文本智能推荐程序还可以被分割为一个或者多个模块,一个或者多个模块被存储于存储器11中,并由一个或多个处理器(本实施例为处理器12)所执行以完成本申请,本申请所称的模块是指能够完成特定功能的一系列计算机程序指令段,用于描述个性化文本智能推荐程序在个性化文本智能推荐装置中的执行过程。
例如,参照图3所示,为本申请个性化文本智能推荐装置一实施例中的个性化文本智能推荐程序的程序模块示意图,该实施例中,所述个性化文本智能推荐程序可以被分割为源数据接收模块10、特征提取模块20、特征分析模块30以及个性化文本输出模块40,示例性地:
所述源数据接收模块10用于:接收包括基础文本数据集和场景文本数据 集的语料集和标签集,将所述语料集进行包括分词、去停用词的预处理操作得到标准语料集。
所述特征提取模块20用于:基于关键字抽取算法对所述标准语料集进行关键字抽取后得到关键字数据集,对所述关键字数据集进行聚类分析得到类别集,将所述类别集与所述关键字数据集进行词向量化操作得到词向量类别集与词向量关键字集。
所述特征分析模块30用于:将所述词向量关键字集输入至推荐模型,所述推荐模型将所述词向量关键字集输入至卷积神经网络中进行训练并得到第一训练值和第二训练值,将所述第一训练值和所述词向量类别集输入至损失函数中,所述损失函数计算得到第一损失值,判断所述第一损失值与预设第一阈值的大小,若所述第一损失值大于预设第一阈值,则所述卷积神经网络继续训练,若所述第一损失值小于预设第一阈值,则将所述第二训练值和所述标签集输入至所述损失函数计算得到第二损失值,判断所述第二损失值与预设第二阈值的大小,若所述第二损失值大于预设第二阈值,所述卷积神经网络继续训练,若所述第二损失值小于预设第二阈值,所述卷积神经网络退出训练。
所述个性化文本输出模块40用于:接收用户输入的文本数据,将所述文本数据输入至所述推荐模型中判断所述文本数据的类别和标签,并根据判断的所述类别和标签从数据库中推荐相同类别和标签的文本内容。
上述源数据接收模块10、特征提取模块20、特征分析模块30以及个性化文本输出模块40等程序模块被执行时所实现的功能或操作步骤与上述实施例大体相同,在此不再赘述。
此外,本申请实施例还提出一种计算机可读存储介质,所述计算机可读存储介质上存储有个性化文本智能推荐程序,所述个性化文本智能推荐程序可被一个或多个处理器执行,以实现如下操作:
接收包括基础文本数据集和场景文本数据集的语料集和标签集,将所述语料集进行包括分词、去停用词的预处理操作得到标准语料集;
基于关键字抽取算法对所述标准语料集进行关键字抽取后得到关键字数据集,对所述关键字数据集进行聚类分析得到类别集,将所述类别集与所述 关键字数据集进行词向量化操作得到词向量类别集与词向量关键字集;
将所述词向量关键字集输入至推荐模型,所述推荐模型将所述词向量关键字集输入至卷积神经网络中进行训练并得到第一训练值和第二训练值,将所述第一训练值和所述词向量类别集输入至损失函数中,所述损失函数计算得到第一损失值,判断所述第一损失值与预设第一阈值的大小,若所述第一损失值大于预设第一阈值,则所述卷积神经网络继续训练,若所述第一损失值小于预设第一阈值,则将所述第二训练值和所述标签集输入至所述损失函数计算得到第二损失值,判断所述第二损失值与预设第二阈值的大小,若所述第二损失值大于预设第二阈值,所述卷积神经网络继续训练,若所述第二损失值小于预设第二阈值,所述卷积神经网络退出训练;
接收用户输入的文本数据,将所述文本数据输入至所述推荐模型中判断所述文本数据的类别和标签,并根据判断的所述类别和标签从数据库中推荐相同类别和标签的文本内容。
本申请计算机可读存储介质具体实施方式与上述个性化文本智能推荐装置和方法各实施例基本相同,在此不作累述。
需要说明的是,上述本申请实施例序号仅仅为了描述,不代表实施例的优劣。并且本文中的术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、装置、物品或者方法不仅包括那些要素,而且还包括没有明确列出的其他要素,或者是还包括为这种过程、装置、物品或者方法所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括该要素的过程、装置、物品或者方法中还存在另外的相同要素。
通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现,当然也可以通过硬件,但很多情况下前者是更佳的实施方式。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来,该计算机软件产品存储在如上所述的一个存储介质(如ROM/RAM、磁碟、光盘)中,包括若干指令用以使得一台终端设备(可以是手机,计算机,服务器,或者网络设备等)执行本申请各个实施例所述的方法。
以上仅为本申请的优选实施例,并非因此限制本申请的专利范围,凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换,或直接或间接运用在其他相关的技术领域,均同理包括在本申请的专利保护范围内。
Claims (20)
- 一种个性化文本智能推荐方法,其特征在于,所述方法包括:接收包括基础文本数据集和场景文本数据集的语料集和标签集,将所述语料集进行包括分词、去停用词的预处理操作得到标准语料集;基于关键字抽取算法对所述标准语料集进行关键字抽取后得到关键字数据集,对所述关键字数据集进行聚类分析得到类别集,将所述类别集与所述关键字数据集进行词向量化操作得到词向量类别集与词向量关键字集;将所述词向量关键字集输入至推荐模型,所述推荐模型将所述词向量关键字集输入至卷积神经网络中进行训练并得到第一训练值和第二训练值,将所述第一训练值和所述词向量类别集输入至损失函数中,所述损失函数计算得到第一损失值,判断所述第一损失值与预设第一阈值的大小,若所述第一损失值大于预设第一阈值,则所述卷积神经网络继续训练,若所述第一损失值小于预设第一阈值,则将所述第二训练值和所述标签集输入至所述损失函数计算得到第二损失值,判断所述第二损失值与预设第二阈值的大小,若所述第二损失值大于预设第二阈值,所述卷积神经网络继续训练,若所述第二损失值小于预设第二阈值,所述卷积神经网络退出训练;接收用户输入的文本数据,将所述文本数据输入至所述推荐模型中判断所述文本数据的类别和标签,并根据判断的所述类别和标签从数据库中推荐相同类别和标签的文本内容。
- 如权利要求1所述的个性化文本智能推荐方法,其特征在于,所述基础文本数据集包括微博评论集、影电观后感集、音乐评论集;所述场景文本数据集包括股票评论集、政府工作报告评论集、公司季度年度财务报表评论集、大学生就业情况评论集。
- 如权利要求1所述的个性化文本智能推荐方法,其特征在于,所述分词包括:根据所述语料集建立概率分词模型P(S)和最大化所述概率分词模型P(S),并利用最大化的所述概率分词模型P(S)对所述语料集执行分词操作;其中,所述概率分词模型P(S)为:其中,W 1,W 2,…,W m为所述语料集包括的词,m为所述语料集的数量, p(W i|W i-1)表示在词W i-1出现的情况下词W i出现的概率;所述最大化的所述概率分词模型P(S):其中,count(W i-1,W i)表示词W i-1和词W i同时出现在所述语料集内同一篇文本的文本数量,count(W i-1)表示词W i-1出现在所述语料集内的文本数量,argmax表示最大化操作。
- 如权利要求1所述的个性化文本智能推荐方法,其特征在于,基于关键字抽取算法对所述标准语料集进行关键字抽取后得到关键字数据集,包括:计算所述标准语料集中任意两词W i,W j之间的依存关联度Dep(W i,W j):其中,len(W i,W j)表示词语W i和W j之间的依存路径长度,b是超参数;计算所述标准语料集中任意两词W i,W j之间的引力值f grav(W i,W j):其中,tfidf(W i)、tfidf(W j)表示词W i,W j的词频-逆文本频率指数,d表示词W i和W j的词向量之间的欧式距离;根据所述依存关联度Dep(W i,W j)和所述引力值f grav(W i,W j)判断所述标准语料集中任意两词W i,W j之间的权重系数weight(W i,W j):weight(W i,W j)=Dep(W i,W j)*f grav(W i,W j)按照所述权重系数大小选择权重系数weight(W i,W j)最大的词,完成所述关键字抽取,得到关键字数据集。
- 如权利要求2所述的个性化文本智能推荐方法,其特征在于,基于关键字抽取算法对所述标准语料集进行关键字抽取后得到关键字数据集,包括:计算所述标准语料集中任意两词W i,W j之间的依存关联度Dep(W i,W j):其中,len(W i,W j)表示词语W i和W j之间的依存路径长度,b是超参数;计算所述标准语料集中任意两词W i,W j之间的引力值f grav(W i,W j):其中,tfidf(W i)、tfidf(W j)表示词W i,W j的词频-逆文本频率指数,d表示词W i和W j的词向量之间的欧式距离;根据所述依存关联度Dep(W i,W j)和所述引力值f grav(W i,W j)判断所述标准 语料集中任意两词W i,W j之间的权重系数weight(W i,W j):weight(W i,W j)=Dep(W i,W j)*f grav(W i,W j)按照所述权重系数大小选择权重系数weight(W i,W j)最大的词,完成所述关键字抽取,得到关键字数据集。
- 如权利要求3所述的个性化文本智能推荐方法,其特征在于,基于关键字抽取算法对所述标准语料集进行关键字抽取后得到关键字数据集,包括:计算所述标准语料集中任意两词W i,W j之间的依存关联度Dep(W i,W j):其中,len(W i,W j)表示词语W i和W j之间的依存路径长度,b是超参数;计算所述标准语料集中任意两词W i,W j之间的引力值f grav(W i,W j):其中,tfidf(W i)、tfidf(W j)表示词W i,W j的词频-逆文本频率指数,d表示词W i和W j的词向量之间的欧式距离;根据所述依存关联度Dep(W i,W j)和所述引力值f grav(W i,W j)判断所述标准语料集中任意两词W i,W j之间的权重系数weight(W i,W j):weight(W i,W j)=Dep(W i,W j)*f grav(W i,W j)按照所述权重系数大小选择权重系数weight(W i,W j)最大的词,完成所述关键字抽取,得到关键字数据集。
- 一种个性化文本智能推荐装置,其特征在于,所述装置包括存储器和处理器,所述存储器上存储有可在所述处理器上运行的个性化文本智能推荐 程序,所述个性化文本智能推荐程序被所述处理器执行时实现如下步骤:接收包括基础文本数据集和场景文本数据集的语料集和标签集,将所述语料集进行包括分词、去停用词的预处理操作得到标准语料集;基于关键字抽取算法对所述标准语料集进行关键字抽取后得到关键字数据集,对所述关键字数据集进行聚类分析得到类别集,将所述类别集与所述关键字数据集进行词向量化操作得到词向量类别集与词向量关键字集;将所述词向量关键字集输入至推荐模型,所述推荐模型将所述词向量关键字集输入至卷积神经网络中进行训练并得到第一训练值和第二训练值,将所述第一训练值和所述词向量类别集输入至损失函数中,所述损失函数计算得到第一损失值,判断所述第一损失值与预设第一阈值的大小,若所述第一损失值大于预设第一阈值,则所述卷积神经网络继续训练,若所述第一损失值小于预设第一阈值,则将所述第二训练值和所述标签集输入至所述损失函数计算得到第二损失值,判断所述第二损失值与预设第二阈值的大小,若所述第二损失值大于预设第二阈值,所述卷积神经网络继续训练,若所述第二损失值小于预设第二阈值,所述卷积神经网络退出训练;接收用户输入的文本数据,将所述文本数据输入至所述推荐模型中判断所述文本数据的类别和标签,并根据判断的所述类别和标签从数据库中推荐相同类别和标签的文本内容。
- 如权利要求8所述的个性化文本智能推荐装置,其特征在于,所述基础文本数据集包括微博评论集、影电观后感集、音乐评论集;所述场景文本数据集包括股票评论集、政府工作报告评论集、公司季度年度财务报表评论集、大学生就业情况评论集。
- 如权利要求8所述的个性化文本智能推荐装置,其特征在于,所述分词包括:根据所述语料集建立概率分词模型P(S)和最大化所述概率分词模型P(S),并利用最大化的所述概率分词模型P(S)对所述语料集执行分词操作;其中,所述概率分词模型P(S)为:其中,W 1,W 2,…,W m为所述语料集包括的词,m为所述语料集的数量,p(W i|W i-1)表示在词W i-1出现的情况下词W i出现的概率;所述最大化的所述概率分词模型P(S):其中,count(W i-1,W i)表示词W i-1和词W i同时出现在所述语料集内同一篇文本的文本数量,count(W i-1)表示词W i-1出现在所述语料集内的文本数量,argmax表示最大化操作。
- 如权利要求8所述的个性化文本智能推荐装置,其特征在于,基于关键字抽取算法对所述标准语料集进行关键字抽取后得到关键字数据集,包括:计算所述标准语料集中任意两词W i,W j之间的依存关联度Dep(W i,W j):其中,len(W i,W j)表示词语W i和W j之间的依存路径长度,b是超参数;计算所述标准语料集中任意两词W i,W j之间的引力值f grav(W i,W j):其中,tfidf(W i)、tfidf(W j)表示词W i,W j的词频-逆文本频率指数,d表示词W i和W j的词向量之间的欧式距离;根据所述依存关联度Dep(W i,W j)和所述引力值f grav(W i,W j)判断所述标准语料集中任意两词W i,W j之间的权重系数weight(W i,W j):weight(W i,W j)=Dep(W i,W j)*f grav(W i,W j)按照所述权重系数大小选择权重系数weight(W i,W j)最大的词,完成所述关键字抽取,得到关键字数据集。
- 如权利要求9所述的个性化文本智能推荐装置,其特征在于,基于关键字抽取算法对所述标准语料集进行关键字抽取后得到关键字数据集,包括:计算所述标准语料集中任意两词W i,W j之间的依存关联度Dep(W i,W j):其中,len(W i,W j)表示词语W i和W j之间的依存路径长度,b是超参数;计算所述标准语料集中任意两词W i,W j之间的引力值f grav(W i,W j):其中,tfidf(W i)、tfidf(W j)表示词W i,W j的词频-逆文本频率指数,d表 示词W i和W j的词向量之间的欧式距离;根据所述依存关联度Dep(W i,W j)和所述引力值f grav(W i,W j)判断所述标准语料集中任意两词W i,W j之间的权重系数weight(W i,W j):weight(W i,W j)=Dep(W i,W j)*f grav(W i,W j)按照所述权重系数大小选择权重系数weight(W i,W j)最大的词,完成所述关键字抽取,得到关键字数据集。
- 如权利要求10所述的个性化文本智能推荐装置,其特征在于,基于关键字抽取算法对所述标准语料集进行关键字抽取后得到关键字数据集,包括:计算所述标准语料集中任意两词W i,W j之间的依存关联度Dep(W i,W j):其中,len(W i,W j)表示词语W i和W j之间的依存路径长度,b是超参数;计算所述标准语料集中任意两词W i,W j之间的引力值f grav(W i,W j):其中,tfidf(W i)、tfidf(W j)表示词W i,W j的词频-逆文本频率指数,d表示词W i和W j的词向量之间的欧式距离;根据所述依存关联度Dep(W i,W j)和所述引力值f grav(W i,W j)判断所述标准语料集中任意两词W i,W j之间的权重系数weight(W i,W j):weight(W i,W j)=Dep(W i,W j)*f grav(W i,W j)按照所述权重系数大小选择权重系数weight(W i,W j)最大的词,完成所述关键字抽取,得到关键字数据集。
- 一种计算机可读存储介质,其特征在于,所述计算机可读存储介质上存储有个性化文本智能推荐程序,所述个性化文本智能推荐程序可被一个或者多个处理器执行,以实现如下步骤:接收包括基础文本数据集和场景文本数据集的语料集和标签集,将所述语料集进行包括分词、去停用词的预处理操作得到标准语料集;基于关键字抽取算法对所述标准语料集进行关键字抽取后得到关键字数据集,对所述关键字数据集进行聚类分析得到类别集,将所述类别集与所述关键字数据集进行词向量化操作得到词向量类别集与词向量关键字集;将所述词向量关键字集输入至推荐模型,所述推荐模型将所述词向量关键字集输入至卷积神经网络中进行训练并得到第一训练值和第二训练值,将所述第一训练值和所述词向量类别集输入至损失函数中,所述损失函数计算得到第一损失值,判断所述第一损失值与预设第一阈值的大小,若所述第一损失值大于预设第一阈值,则所述卷积神经网络继续训练,若所述第一损失值小于预设第一阈值,则将所述第二训练值和所述标签集输入至所述损失函数计算得到第二损失值,判断所述第二损失值与预设第二阈值的大小,若所述第二损失值大于预设第二阈值,所述卷积神经网络继续训练,若所述第二损失值小于预设第二阈值,所述卷积神经网络退出训练;接收用户输入的文本数据,将所述文本数据输入至所述推荐模型中判断所述文本数据的类别和标签,并根据判断的所述类别和标签从数据库中推荐相同类别和标签的文本内容。
- 如权利要求15所述的计算机可读存储介质,其特征在于,所述基础文本数据集包括微博评论集、影电观后感集、音乐评论集;所述场景文本数据集包括股票评论集、政府工作报告评论集、公司季度年度财务报表评论集、大学生就业情况评论集。
- 如权利要求15所述的个性化文本智能推荐装置,其特征在于,所述分词包括:根据所述语料集建立概率分词模型P(S)和最大化所述概率分词模型P(S),并利用最大化的所述概率分词模型P(S)对所述语料集执行分词操作;其中,所述概率分词模型P(S)为:其中,W 1,W 2,…,W m为所述语料集包括的词,m为所述语料集的数量,p(W i|W i-1)表示在词W i-1出现的情况下词W i出现的概率;所述最大化的所述概率分词模型P(S):其中,count(W i-1,W i)表示词W i-1和词W i同时出现在所述语料集内同一篇文本的文本数量,count(W i-1)表示词W i-1出现在所述语料集内的文本数量,argmax表示最大化操作。
- 如权利要求15所述的计算机可读存储介质,其特征在于,基于关键字抽取算法对所述标准语料集进行关键字抽取后得到关键字数据集,包括:计算所述标准语料集中任意两词W i,W j之间的依存关联度Dep(W i,W j):其中,len(W i,W j)表示词语W i和W j之间的依存路径长度,b是超参数;计算所述标准语料集中任意两词W i,W j之间的引力值f grav(W i,W j):其中,tfidf(W i)、tfidf(W j)表示词W i,W j的词频-逆文本频率指数,d表示词W i和W j的词向量之间的欧式距离;根据所述依存关联度Dep(W i,W j)和所述引力值f grav(W i,W j)判断所述标准语料集中任意两词W i,W j之间的权重系数weight(W i,W j):weight(W i,W j)=Dep(W i,W j)*f grav(W i,W j)按照所述权重系数大小选择权重系数weight(W i,W j)最大的词,完成所述关键字抽取,得到关键字数据集。
- 如权利要求16或17所述的计算机可读存储介质,其特征在于,基于关键字抽取算法对所述标准语料集进行关键字抽取后得到关键字数据集,包括:计算所述标准语料集中任意两词W i,W j之间的依存关联度Dep(W i,W j):其中,len(W i,W j)表示词语W i和W j之间的依存路径长度,b是超参数;计算所述标准语料集中任意两词W i,W j之间的引力值f grav(W i,W j):其中,tfidf(W i)、tfidf(W j)表示词W i,W j的词频-逆文本频率指数,d表示词W i和W j的词向量之间的欧式距离;根据所述依存关联度Dep(W i,W j)和所述引力值f grav(W i,W j)判断所述标准语料集中任意两词W i,W j之间的权重系数weight(W i,W j):weight(W i,W j)=Dep(W i,W j)*f grav(W i,W j)按照所述权重系数大小选择权重系数weight(W i,W j)最大的词,完成所述关键字抽取,得到关键字数据集。
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910582849.7 | 2019-06-28 | ||
| CN201910582849.7A CN110427480B (zh) | 2019-06-28 | 2019-06-28 | 个性化文本智能推荐方法、装置及计算机可读存储介质 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2020258481A1 true WO2020258481A1 (zh) | 2020-12-30 |
Family
ID=68408912
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2019/102201 Ceased WO2020258481A1 (zh) | 2019-06-28 | 2019-08-23 | 个性化文本智能推荐方法、装置及计算机可读存储介质 |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN110427480B (zh) |
| WO (1) | WO2020258481A1 (zh) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112989040A (zh) * | 2021-03-10 | 2021-06-18 | 河南中原消费金融股份有限公司 | 一种对话文本标注方法、装置、电子设备及存储介质 |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111275091B (zh) * | 2020-01-16 | 2024-05-10 | 平安科技(深圳)有限公司 | 文本结论智能推荐方法、装置及计算机可读存储介质 |
| CN111737456B (zh) * | 2020-05-15 | 2024-08-20 | 恩亿科(北京)数据科技有限公司 | 一种语料信息的处理方法和装置 |
| CN112184465B (zh) * | 2020-09-27 | 2024-09-17 | 中国平安财产保险股份有限公司 | 用户数据画像生成方法、装置、电子设备及存储介质 |
| CN112559740A (zh) * | 2020-12-03 | 2021-03-26 | 星宏传媒有限公司 | 一种基于多模型融合的广告标签分类方法、系统及设备 |
| CN115982429B (zh) * | 2023-03-21 | 2023-08-01 | 中交第四航务工程勘察设计院有限公司 | 一种基于流程控制的知识管理方法及系统 |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104298732A (zh) * | 2014-09-29 | 2015-01-21 | 中国科学院计算技术研究所 | 一种面向网络用户的个性化文本排序及推荐方法 |
| US20150227589A1 (en) * | 2014-02-10 | 2015-08-13 | Microsoft Corporation | Semantic matching and annotation of attributes |
| CN106776881A (zh) * | 2016-11-28 | 2017-05-31 | 中国科学院软件研究所 | 一种基于微博平台的领域信息推荐系统及方法 |
| CN107315797A (zh) * | 2017-06-19 | 2017-11-03 | 江西洪都航空工业集团有限责任公司 | 一种网络新闻获取及文本情感预测系统 |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110137898A1 (en) * | 2009-12-07 | 2011-06-09 | Xerox Corporation | Unstructured document classification |
| US20140207716A1 (en) * | 2013-01-22 | 2014-07-24 | Maluuba Inc. | Natural language processing method and system |
| WO2015106657A1 (zh) * | 2014-01-16 | 2015-07-23 | 上海资本加管理软件有限公司 | 一种应用于社交网络的推荐方法和推荐系统 |
| CN105677769B (zh) * | 2015-12-29 | 2018-01-05 | 广州神马移动信息科技有限公司 | 一种基于潜在狄利克雷分配(lda)模型的关键词推荐方法和系统 |
| US10489438B2 (en) * | 2016-05-19 | 2019-11-26 | Conduent Business Services, Llc | Method and system for data processing for text classification of a target domain |
| CN108319627B (zh) * | 2017-02-06 | 2024-05-28 | 腾讯科技(深圳)有限公司 | 关键词提取方法以及关键词提取装置 |
| CN108573047A (zh) * | 2018-04-18 | 2018-09-25 | 广东工业大学 | 一种中文文本分类模型的训练方法及装置 |
| CN109299270B (zh) * | 2018-10-30 | 2021-09-28 | 云南电网有限责任公司信息中心 | 一种基于卷积神经网络的文本数据无监督聚类方法 |
| CN109857860A (zh) * | 2019-01-04 | 2019-06-07 | 平安科技(深圳)有限公司 | 文本分类方法、装置、计算机设备及存储介质 |
-
2019
- 2019-06-28 CN CN201910582849.7A patent/CN110427480B/zh active Active
- 2019-08-23 WO PCT/CN2019/102201 patent/WO2020258481A1/zh not_active Ceased
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150227589A1 (en) * | 2014-02-10 | 2015-08-13 | Microsoft Corporation | Semantic matching and annotation of attributes |
| CN104298732A (zh) * | 2014-09-29 | 2015-01-21 | 中国科学院计算技术研究所 | 一种面向网络用户的个性化文本排序及推荐方法 |
| CN106776881A (zh) * | 2016-11-28 | 2017-05-31 | 中国科学院软件研究所 | 一种基于微博平台的领域信息推荐系统及方法 |
| CN107315797A (zh) * | 2017-06-19 | 2017-11-03 | 江西洪都航空工业集团有限责任公司 | 一种网络新闻获取及文本情感预测系统 |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112989040A (zh) * | 2021-03-10 | 2021-06-18 | 河南中原消费金融股份有限公司 | 一种对话文本标注方法、装置、电子设备及存储介质 |
| CN112989040B (zh) * | 2021-03-10 | 2024-02-27 | 河南中原消费金融股份有限公司 | 一种对话文本标注方法、装置、电子设备及存储介质 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110427480B (zh) | 2022-10-11 |
| CN110427480A (zh) | 2019-11-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113434636B (zh) | 基于语义的近似文本搜索方法、装置、计算机设备及介质 | |
| US11657231B2 (en) | Capturing rich response relationships with small-data neural networks | |
| US11030199B2 (en) | Systems and methods for contextual retrieval and contextual display of records | |
| US10643109B2 (en) | Method and system for automatically classifying data expressed by a plurality of factors with values of text word and symbol sequence by using deep learning | |
| CN110442857B (zh) | 情感智能判断方法、装置及计算机可读存储介质 | |
| CN110765765B (zh) | 基于人工智能的合同关键条款提取方法、装置及存储介质 | |
| US9767144B2 (en) | Search system with query refinement | |
| WO2020258481A1 (zh) | 个性化文本智能推荐方法、装置及计算机可读存储介质 | |
| WO2020237856A1 (zh) | 基于知识图谱的智能问答方法、装置及计算机存储介质 | |
| US20210097238A1 (en) | User keyword extraction device and method, and computer-readable storage medium | |
| US9934293B2 (en) | Generating search results | |
| WO2021068339A1 (zh) | 文本分类方法、装置及计算机可读存储介质 | |
| US10599711B1 (en) | Region-specific image download probability modeling | |
| WO2020000717A1 (zh) | 网页分类方法、装置及计算机可读存储介质 | |
| CN114116997A (zh) | 知识问答方法、装置、电子设备及存储介质 | |
| CN112182348B (zh) | 语义匹配判定方法、装置、电子设备、计算机可读介质 | |
| WO2020253043A1 (zh) | 智能文本分类方法、装置及计算机可读存储介质 | |
| US12517970B2 (en) | Method and system for providing alternative result for an online search previously with no result | |
| CN113609847B (zh) | 信息抽取方法、装置、电子设备及存储介质 | |
| CN113282763B (zh) | 文本关键信息提取、装置、设备及存储介质 | |
| CN116894076A (zh) | 生成用于机器学习的基于用户的训练数据 | |
| WO2021042529A1 (zh) | 文章摘要自动生成方法、装置及计算机可读存储介质 | |
| CN115098619A (zh) | 资讯去重方法、装置、电子设备及计算机可读取存储介质 | |
| Chou et al. | On the construction of web ner model training tool based on distant supervision | |
| CN114329206A (zh) | 标题生成方法和装置、电子设备、计算机可读介质 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 19934834 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 19934834 Country of ref document: EP Kind code of ref document: A1 |









