CN109933773B - Multiple semantic statement analysis system and method - Google Patents
Multiple semantic statement analysis system and method Download PDFInfo
- Publication number
- CN109933773B CN109933773B CN201711353550.1A CN201711353550A CN109933773B CN 109933773 B CN109933773 B CN 109933773B CN 201711353550 A CN201711353550 A CN 201711353550A CN 109933773 B CN109933773 B CN 109933773B
- Authority
- CN
- China
- Prior art keywords
- neural network
- text
- vectors
- semantic
- feature
- 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
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Machine Translation (AREA)
Abstract
The invention relates to a multiple semantic sentence analysis system which comprises a text encoding module, a neural network module and a text decoding module, wherein the text encoding module converts text information into a plurality of groups of text vector information, the neural network module converts the plurality of groups of text vector information into a plurality of groups of semantic vectors according to a set algorithm, and the text decoding module converts the plurality of groups of semantic vectors into text information and outputs the text information. The invention also discloses a multi-semantic sentence analysis method, which comprises the steps of text coding, text logic analysis, text decoding output and the like. The invention adopts the neural network module containing a plurality of neural network algorithms to analyze the multiple semantic sentences, and has the characteristic of being capable of accurately and efficiently analyzing the multiple semantic sentences.
Description
Technical Field
The invention relates to a semantic sentence analysis system and a semantic sentence analysis method, in particular to a semantic sentence analysis system and a semantic sentence analysis method with multiple semantic sentences, and belongs to the field of natural language artificial intelligence.
Background
With the continuous perfection and development of neural network technology, artificial intelligent products gradually enter various fields, and products with man-machine interaction functions are more and more favored by people. The intelligent product can complete man-machine interaction, understand the intention of the user and then make corresponding actions. For example, the apple Siri company, the user can wake it up and tell it to make a call, listen to music, etc., and then the electronic product will automatically complete these instructions. Microsoft ice can be used to simply chat with the user. The man-machine interaction function of the products saves the time of users (without hands, direct speaking) on one hand, and increases the interest (chat interaction) of the products on the other hand. In the back of achieving these functions, semantic understanding is a critical technology and is also a technology to be improved.
Current semantic understanding techniques are capable of accurately understanding sentences that contain only a single semantic meaning, e.g., users: navigating to the people square. And (3) a robot: is navigating to people squares. However, for sentences containing multiple semantics, such as users: the user needs two instructions executed by the robot to navigate to the people square and eat meal at the affordable restaurant, namely, navigate to the people square and search for the affordable restaurant near the people square. This type of statement contains two or more semantics, which cannot be resolved accurately and efficiently by current semantic understanding techniques.
Disclosure of Invention
The invention discloses a novel scheme for analyzing multiple semantic sentences by adopting a neural network module comprising a plurality of neural network algorithms, and solves the problem that the conventional scheme cannot accurately and efficiently analyze the multiple semantic sentences.
The multi-semantic sentence analysis system comprises a text encoding module, a neural network module and a text decoding module, wherein the text encoding module converts text information into a plurality of groups of text vector information, the neural network module converts the plurality of groups of text vector information into a plurality of groups of semantic vectors according to a set algorithm, and the text decoding module converts the plurality of groups of semantic vectors into text information and outputs the text information.
The invention also discloses a multiple semantic sentence analysis method, which is based on a multiple semantic sentence analysis system, wherein the multiple semantic sentence analysis system comprises a text encoding module, a neural network module and a text decoding module, and comprises the following steps: the text encoding module performs text encoding on input text information and then converts the text information into a word vector format; the neural network module breaks down text information in a word vector format into a plurality of sentences in the word vector format; thirdly, the neural network module carries out logic analysis and calculation on the sentences in the multiple word vector formats according to a set algorithm to obtain multiple groups of semantic vectors; and the text decoding module decodes the texts by the plurality of groups of semantic vectors, converts the text information into text information and outputs the text information to a specific application layer.
Further, the processing procedure of the neural network module of the method of the present solution includes: the method comprises the steps of processing an input sentence in a word vector format by adopting a convolutional neural network, extracting a feature vector representing the sentence, and adjusting the feature vector according to an application scene; secondly, decomposing the feature vectors into a corresponding number of feature vectors according to the number of the semantics contained in the sentences by adopting a cyclic neural network, wherein the decomposed feature vectors correspond to the semantics contained in the sentences; thirdly, the cyclic neural network and the deep neural network are combined to respectively carry out logic operation on the decomposed feature vectors to extract semantic vectors.
Furthermore, in the processing procedure of the neural network module of the method of the scheme, the cyclic neural network and the deep neural network are combined to respectively carry out logic operation on the decomposed feature vectors to obtain semantic vectors and logic vectors, the logic vectors represent the dependency relationship among the semantics, and the subsequent semantic vectors are obtained by combining the corresponding feature vectors with the logic vectors of the last feature vector.
The multiple semantic sentence analysis system and the multiple semantic sentence analysis method adopt the neural network module comprising a plurality of neural network algorithms to analyze the multiple semantic sentences, and have the characteristic of being capable of accurately and efficiently analyzing the multiple semantic sentences.
Drawings
FIG. 1 is a schematic diagram of a multiple semantic statement parsing system and method of the present invention.
Fig. 2 is a flow chart of the neural network module processing information.
Detailed Description
The invention discloses a method for understanding natural language multiple semantics by adopting artificial intelligence. The current deep learning technology has great success in the fields of text classification, text labeling, translation and the like, so the scheme provides a semantic understanding method based on the deep learning technology, which is used for solving the problem of multiple semantic sentence analysis. As shown in fig. 1, the multiple semantic understanding method of the present scheme is a block diagram. The scheme comprises three modules, namely a text encoding module, a neural network module and a text decoding module. The text coding module is a module capable of converting a text into a calculation, the text is coded in a word embedding mode, after the step, the text is coded into a plurality of groups of vectors, the step is also called vectorization of the text, and a large number of texts are needed to train the model in advance. The neural network module is a core module of the present solution, and the neural network model is generally built in an end-to-end (end to end) manner, that is, input corresponds to output. The method adds a logical reasoning processing method on an end-to-end basis, calculates the logical relation between the current sentence and the first several sentences, so that the model can be combined with the context to comprehensively understand multiple semantics in the input text. The text decoding module is similar to the text encoding module, and is used for converting the output vector of the neural network into specific words, so that the model also needs to be trained by using a large amount of texts in advance.
As shown in fig. 1, taking "navigate to people square and eat rice" as an example, the multiple semantic method of the present solution includes the following processing steps:
the method includes the steps of text encoding and converting the text encoding into word vectors.
The neural network module analyzes the text, and decomposes the text into a first sentence: navigating to get to people squares; statement two: and eat meal.
Thirdly, the neural network module carries out logic analysis and calculation on the first statement and the second statement, and outputs the first semantic meaning: navigation, people squares; semantic two: searching for restaurants and people in the vicinity of squares.
And decoding the text and outputting the decoded text to a specific application layer.
Fig. 2 shows a neural network model block diagram of the present solution. The input of the model is the coded text (word vector), and the output is the semantic vector after the text is analyzed. The algorithm used by the neural network model of the scheme is as follows: convolutional neural network (Convolutional Neural Network, CNN), cyclic neural network (Recurrent Neural Networks, RNN), deep neural network (Deep Neural Network, DNN).
CNN is a multi-layer neural network used to process two-dimensional data, and CNN is considered to be the first truly successful robust deep learning method employing a multi-layer hierarchical network. CNN reduces the number of trainable parameters in the network by mining spatial correlations in the data, achieving improved back-propagation algorithm efficiency for forward-propagation networks, which is also considered a method of deep learning because CNN requires very little data preprocessing effort.
The RNN is used to process sequence data, i.e. one sequence current output is related to the previous output. The specific expression is that the network will memorize the previous information and apply it to the calculation of the current output, i.e. the nodes between the hidden layers are no longer connectionless but connected, and the input of the hidden layer includes not only the output of the input layer but also the output of the hidden layer at the previous moment. In theory, RNNs can process sequence data of any length, thus solving the problem that a common neural network is fully connected from an input layer to an implied layer to an output layer, but nodes between layers are connectionless, for example, you want to predict what the next word of a sentence is, and it is generally necessary to use the previous word because the words before and after in a sentence are not independent. RNNs have proven to be very successful in practice for word vector expressions, statement legitimacy checks, part-of-speech tagging, etc. Among RNNs, the most widely used and successful model at present is LSTMs (Long Short-Term Memory model) which can better express Long-Short Term dependency.
DNN refers to deep neural network, and is different from RNN recurrent neural network and CNN convolutional neural network in that DNN refers to fully-connected neuron structure, and layers are fully-connected and do not contain convolutional units or are associated in time, that is, any neuron of the i layer must be connected with any neuron of the i+1th layer. DNN can also be understood as a neural network with many hidden layers.
Taking the example of processing "navigate to people square, eat meal, then watch movie", assume that this sentence has been processed into a word vector, the word vector size is (R 1 ,F 1 ),R 1 Representing the number of words, F 1 Representing the length of the word vector, the main steps of the model include the following.
First, the input word vector is processed with CNN, and feature vectors that can represent the sentence are extracted. The size of the feature vector can be customized, and generally needs to be adjusted according to the application scene. The processed feature vector has a size (R 2 ,F 2 ),R 2 Represents the length, F 2 Representing the feature number.
Then, based on the number of semantics contained in the sentence, the feature vector is decomposed by RNN, and the sentence contains three kinds of semanticsSemantic, and therefore, decomposed into feature vectors 1, of size (N 1 ,F 2 ) Feature vector 2, size (N 2 ,F 2 ) Feature vector 3, size (N 3 ,F 2 ) The three feature vectors correspond to three semantics of navigation, eating and film watching in sentences respectively.
Finally, combining RNN and DNN, and respectively carrying out logic operation on the decomposed feature vectors to extract semantic vectors. In order for the neural network to "remember" the logical relationships between the semantics, the present model adds a logical vector to each semantic, which is used to represent the dependency of the present semantic with the previous semantic. In this example, for the first semantic vector, since there is no text in front of it, only feature vector 1 is used to calculate to obtain semantic vector 1 and logic vector 1, and the subsequent semantic vector is combined by the feature vector and the logic vector of the previous semantic to obtain the result.
The model finally outputs semantic vectors, and specific semantics can be obtained after decoding, so that the semantic vectors are provided for upper-layer applications to react to instructions of users.
The multi-semantic statement analysis system and the multi-semantic statement analysis method of the scheme combine the advanced deep learning technology, comprehensively use the CNN, DNN, RNN technology and provide a method capable of accurately analyzing multi-semantics. Practice proves that the method can effectively solve the problem of multi-semantic sentence analysis in man-machine interaction after large-scale corpus training. Based on the characteristics, the multi-semantic statement analysis system and the multi-semantic statement analysis method have outstanding substantive characteristics and remarkable progress compared with the existing scheme.
The system and method for parsing multiple semantic sentences in the present solution are not limited to those disclosed in the specific embodiments, and the technical solutions presented in the examples may be extended based on understanding of those skilled in the art, and simple alternatives made by those skilled in the art according to the present solution in combination with common general knowledge also belong to the scope of the present solution.
Claims (2)
1. The multi-semantic sentence analysis system is characterized by comprising a text encoding module, a neural network module and a text decoding module, wherein the text encoding module converts text information into a plurality of groups of text vector information, the neural network module converts the plurality of groups of text vector information into a plurality of groups of semantic vectors according to a set algorithm, and the text decoding module converts the plurality of groups of semantic vectors into text information and outputs the text information; the processing process of the neural network module comprises the following steps:
the method comprises the steps of processing an input sentence in a word vector format by adopting a convolutional neural network, extracting a feature vector representing the sentence, and customizing the size of the feature vector according to an application scene, wherein the size of the processed feature vector is (R2, F2), R2 represents the length, and F2 represents the feature number;
secondly, decomposing the feature vectors into a corresponding number of feature vectors according to the number of the semantics contained in the sentences by adopting a cyclic neural network, wherein the decomposed feature vectors correspond to the semantics contained in the sentences;
thirdly, respectively carrying out logic operation on the decomposed feature vectors by adopting the combination of a cyclic neural network and a deep neural network to extract semantic vectors;
in the processing procedure of the neural network module, a cyclic neural network and a deep neural network are combined to respectively carry out logic operation on the decomposed feature vectors to obtain semantic vectors and logic vectors, the logic vectors represent the dependency relationship among semantics, and the subsequent semantic vectors are obtained by combining the corresponding feature vectors with the logic vectors of the last feature vector.
2. The multi-semantic sentence analysis method is based on a multi-semantic sentence analysis system, and the multi-semantic sentence analysis system comprises a text encoding module, a neural network module and a text decoding module and is characterized by comprising the following steps:
the text encoding module performs text encoding on input text information and then converts the text information into a word vector format;
the neural network module breaks down text information in a word vector format into a plurality of sentences in the word vector format;
thirdly, the neural network module carries out logic analysis and calculation on the sentences in the multiple word vector formats according to a set algorithm to obtain multiple groups of semantic vectors;
the text decoding module decodes the text of the plurality of groups of semantic vectors into text information and outputs the text information to a specific application layer;
the processing process of the neural network module comprises the following steps:
the method comprises the steps of processing an input sentence in a word vector format by adopting a convolutional neural network, extracting a feature vector representing the sentence, and customizing the size of the feature vector according to an application scene, wherein the size of the processed feature vector is (R2, F2), R2 represents the length, and F2 represents the feature number;
secondly, decomposing the feature vectors into a corresponding number of feature vectors according to the number of the semantics contained in the sentences by adopting a cyclic neural network, wherein the decomposed feature vectors correspond to the semantics contained in the sentences;
thirdly, respectively carrying out logic operation on the decomposed feature vectors by adopting the combination of a cyclic neural network and a deep neural network to extract semantic vectors;
in the processing procedure of the neural network module, a cyclic neural network and a deep neural network are combined to respectively carry out logic operation on the decomposed feature vectors to obtain semantic vectors and logic vectors, the logic vectors represent the dependency relationship among semantics, and the subsequent semantic vectors are obtained by combining the corresponding feature vectors with the logic vectors of the last feature vector.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711353550.1A CN109933773B (en) | 2017-12-15 | 2017-12-15 | Multiple semantic statement analysis system and method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711353550.1A CN109933773B (en) | 2017-12-15 | 2017-12-15 | Multiple semantic statement analysis system and method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109933773A CN109933773A (en) | 2019-06-25 |
CN109933773B true CN109933773B (en) | 2023-05-26 |
Family
ID=66980529
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711353550.1A Active CN109933773B (en) | 2017-12-15 | 2017-12-15 | Multiple semantic statement analysis system and method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109933773B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11157705B2 (en) * | 2019-07-22 | 2021-10-26 | International Business Machines Corporation | Semantic parsing using encoded structured representation |
CN112052953B (en) * | 2020-07-21 | 2022-09-09 | 清华大学 | Embeddable cascade logic system for neural inference system and inference method thereof |
CN113239166B (en) * | 2021-05-24 | 2023-06-06 | 清华大学深圳国际研究生院 | Automatic man-machine interaction method based on semantic knowledge enhancement |
CN114021576A (en) * | 2021-10-27 | 2022-02-08 | 四川启睿克科技有限公司 | Text-based natural language understanding decision method |
CN114881040B (en) * | 2022-05-12 | 2022-12-06 | 桂林电子科技大学 | Method and device for processing semantic information of paragraphs and storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105787560A (en) * | 2016-03-18 | 2016-07-20 | 北京光年无限科技有限公司 | Dialogue data interaction processing method and device based on recurrent neural network |
CN106847271A (en) * | 2016-12-12 | 2017-06-13 | 北京光年无限科技有限公司 | A kind of data processing method and device for talking with interactive system |
CN106875940A (en) * | 2017-03-06 | 2017-06-20 | 吉林省盛创科技有限公司 | A kind of Machine self-learning based on neutral net builds knowledge mapping training method |
CN107153672A (en) * | 2017-03-22 | 2017-09-12 | 中国科学院自动化研究所 | User mutual intension recognizing method and system based on Speech Act Theory |
US9830315B1 (en) * | 2016-07-13 | 2017-11-28 | Xerox Corporation | Sequence-based structured prediction for semantic parsing |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10769189B2 (en) * | 2015-11-13 | 2020-09-08 | Microsoft Technology Licensing, Llc | Computer speech recognition and semantic understanding from activity patterns |
US9858263B2 (en) * | 2016-05-05 | 2018-01-02 | Conduent Business Services, Llc | Semantic parsing using deep neural networks for predicting canonical forms |
-
2017
- 2017-12-15 CN CN201711353550.1A patent/CN109933773B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105787560A (en) * | 2016-03-18 | 2016-07-20 | 北京光年无限科技有限公司 | Dialogue data interaction processing method and device based on recurrent neural network |
US9830315B1 (en) * | 2016-07-13 | 2017-11-28 | Xerox Corporation | Sequence-based structured prediction for semantic parsing |
CN106847271A (en) * | 2016-12-12 | 2017-06-13 | 北京光年无限科技有限公司 | A kind of data processing method and device for talking with interactive system |
CN106875940A (en) * | 2017-03-06 | 2017-06-20 | 吉林省盛创科技有限公司 | A kind of Machine self-learning based on neutral net builds knowledge mapping training method |
CN107153672A (en) * | 2017-03-22 | 2017-09-12 | 中国科学院自动化研究所 | User mutual intension recognizing method and system based on Speech Act Theory |
Non-Patent Citations (2)
Title |
---|
利用深度去噪自编码器深度学习的指令意图理解方法;李瀚清等;《上海交通大学学报》;20160728(第07期);第118-123页 * |
基于深度学习的问答匹配方法;荣光辉等;《计算机应用》;20171010(第10期);第133-137页 * |
Also Published As
Publication number | Publication date |
---|---|
CN109933773A (en) | 2019-06-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109933773B (en) | Multiple semantic statement analysis system and method | |
US11194972B1 (en) | Semantic sentiment analysis method fusing in-depth features and time sequence models | |
CN108763284B (en) | Question-answering system implementation method based on deep learning and topic model | |
CN111639175B (en) | Self-supervision dialogue text abstract method and system | |
CN111581401B (en) | Local citation recommendation system and method based on depth correlation matching | |
CN110209789B (en) | Multi-modal dialog system and method for guiding user attention | |
US20230244879A1 (en) | Language representation model system, pre-training method and apparatus, device, and medium | |
CN110717017A (en) | Method for processing corpus | |
CN113127624B (en) | Question-answer model training method and device | |
CN110059324B (en) | Neural network machine translation method and device based on dependency information supervision | |
CN111581970B (en) | Text recognition method, device and storage medium for network context | |
CN110245228A (en) | The method and apparatus for determining text categories | |
CN116204674B (en) | Image description method based on visual concept word association structural modeling | |
CN113065344A (en) | Cross-corpus emotion recognition method based on transfer learning and attention mechanism | |
CN112100375A (en) | Text information generation method and device, storage medium and equipment | |
CN113705315A (en) | Video processing method, device, equipment and storage medium | |
CN108363685B (en) | Self-media data text representation method based on recursive variation self-coding model | |
CN116050352A (en) | Text encoding method and device, computer equipment and storage medium | |
Nithya et al. | Deep learning based analysis on code-mixed tamil text for sentiment classification with pre-trained ulmfit | |
CN114611529B (en) | Intention recognition method and device, electronic equipment and storage medium | |
CN116341564A (en) | Problem reasoning method and device based on semantic understanding | |
CN114519353B (en) | Model training method, emotion message generation method and device, equipment and medium | |
CN112800186B (en) | Reading understanding model training method and device and reading understanding method and device | |
Jalaja et al. | A behavioral chatbot using encoder-decoder architecture: Humanizing conversations | |
CN114648005A (en) | Multi-fragment machine reading understanding method and device for multitask joint learning |
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 |