CN113626614B - Method, device, equipment and storage medium for constructing information text generation model - Google Patents
Method, device, equipment and storage medium for constructing information text generation model Download PDFInfo
- Publication number
- CN113626614B CN113626614B CN202110955326.XA CN202110955326A CN113626614B CN 113626614 B CN113626614 B CN 113626614B CN 202110955326 A CN202110955326 A CN 202110955326A CN 113626614 B CN113626614 B CN 113626614B
- Authority
- CN
- China
- Prior art keywords
- information
- key attribute
- clause
- article
- text generation
- 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
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
- G06F40/295—Named entity recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/02—Knowledge representation; Symbolic representation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Animal Behavior & Ethology (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Machine Translation (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a construction method, a device, equipment and a storage medium of an information text generation model, wherein the method comprises the following steps: acquiring information article information, extracting key attribute information of the information article information, and constructing an information knowledge graph; and carrying out graph convolution operation on the information knowledge graph according to the information knowledge graph to obtain an information text generation model. The information knowledge graph constructed by the invention can be used for ensuring that effective external knowledge can be integrated into the information graph in the subsequent information text generation process, and can be used as a hidden variable to guide the decoding calculation process, thereby achieving the purpose of controlling the logical relationship between sentences, simultaneously playing the reasoning capability of the graph neural network and ensuring the fact correctness of the generated article.
Description
Technical Field
The invention relates to the technical field of data processing, in particular to a construction method and device of an information text generation model, electronic equipment and a storage medium.
Background
With the rapid development of knowledge graph and deep learning technologies, the need for rapid landing in industry for text generation based on these leading edge technologies is increasing. The automobile industry knowledge is relatively specialized, and is different from the open field or other knowledge fields: the vehicle type brands are various, the configuration parameters are complex, the measures described in the vehicle information are numerous, such as appearance, power, interior decoration and the like, and meanwhile, the information article categories are numerous, such as new vehicles are marketed, vehicle type public praise comparison, preferential information and the like, and the difficulty of text generation is greatly increased.
In the prior art, the generation of the information text is mostly completed by a simple deep learning model based on graph data to text sequences, or only simple external knowledge is combined, and the graph data is not shown by utilizing a graph neural network and then is used as the external knowledge, so that the generated content of the information text is not rich enough, the sentence of an article is single, the word repetition easily occurs, the sentence is not smooth enough, and the complex scene cannot be coped with.
Therefore, there is a need for a method of constructing an information text generation model that can meet the application requirements of the automotive field and rapidly generate a large amount of high-quality text information data.
Disclosure of Invention
To this end, the present invention provides a method, apparatus, electronic device and storage medium for constructing an information text generation model, in an effort to solve or at least alleviate at least one of the above-mentioned problems.
According to an aspect of the present invention, there is provided a construction method of an information text generation model, the method being adapted to construct the information text generation model using a graph neural network method, the method comprising the steps of: acquiring information article information, extracting key attribute information of the information article information, and constructing an information knowledge graph; and carrying out graph convolution operation on the information knowledge graph according to the information knowledge graph to obtain an information text generation model.
Optionally, the step of obtaining information article information, extracting key attribute information of the information article information, and constructing an information knowledge graph includes: acquiring information article information and classifying according to class labels of the information article information; extracting key attribute information from the information article information to obtain the key attribute information of the information article information; according to the key attribute information of the information article information, performing word segmentation and sentence segmentation operation on the information article information; acquiring a first clause set corresponding to the key attribute information according to the word segmentation and clause operation, wherein the first clause set comprises clauses corresponding to the key attribute information and the ordering of the key attribute information in the clauses; acquiring a second clause set according to the first clause set, wherein the second clause set comprises global clauses formed by elements of all the first clause sets and key attribute information contexts corresponding to the clauses; and constructing an information knowledge graph according to the first sentence set and the second sentence set.
Optionally, the step of extracting the key attribute information of the information article information to obtain the key attribute information of the information article information includes: cleaning the information article information to remove useless information in the information article information; extracting keywords from the cleaned information article information to obtain a first key attribute set; performing main-predicate-guest triple extraction and causal triple extraction on the cleaned information article information, and performing de-duplication treatment on the extraction result to obtain a second key attribute set; carrying out automobile domain entity identification on the cleaned information article information to obtain a third key attribute set; and acquiring the key attribute information of the information article information according to the intersection of the elements in the first key attribute set, the second key attribute set and the third key attribute set.
Optionally, the step of extracting key attribute information of the information article information and obtaining the key attribute information of the information article information further includes: and according to the classification of the information article information, compiling a regular expression to extract unique key attribute elements under the classification of the information article information, and taking the key attribute elements as supplements of the key attribute information.
Optionally, the step of obtaining a first clause set corresponding to the key attribute information according to the word segmentation and the clause operation, where the first clause set includes a clause corresponding to the key attribute information, and the step of ordering the key attribute information in the clause includes: performing word segmentation and sentence segmentation operation on the information article information, and extracting sentences corresponding to the key attribute information in the information article information; sorting the key attribute information according to the sequence of occurrence of the key attribute information in the clause, and obtaining a fourth key attribute set; and acquiring a first clause set according to the clause and the fourth key attribute set.
Optionally, the step of obtaining the second clause set according to the first clause set includes: according to a first clause set, acquiring a global clause formed by all elements in the first clause set; acquiring a corresponding key attribute information context of the global clause according to the global clause formed by all elements in the first clause set; and acquiring a second clause set according to the global clause and the corresponding key attribute information context of the global clause.
Optionally, the step of performing a graph convolution operation on the information knowledge graph according to the information knowledge graph to obtain an information text generation model includes: carrying out knowledge representation on the information knowledge graph by using a graph convolution network algorithm to obtain first representation global context information; constructing a text generation network, and coding each piece of key attribute information to obtain a first coding set; connecting the first coding set by using a multi-layer neural network, and calculating to obtain a first plan decoder; acquiring a first operation output according to the first representation global context information and the first plan decoder, wherein an operation result at any moment in the first operation output is related to an operation process at the next moment; acquiring a first hidden variable at any moment, wherein the first hidden variable at any moment is obtained by the first operation output, the first representation global context information and a first hidden variable at the previous moment through a variational self-encoder; according to the first representation global context information, performing sentence-level decoding operation on the first hidden variable at any moment, taking a decoding result as calculation input of the first hidden variable at the next moment, and participating in decoding; and performing iterative training until the loss value converges, and obtaining an information text generation model, wherein the generated text index of the information text generation model is larger than a set threshold value.
According to yet another aspect of the present invention, a method of generating an information text using an information text generation model is disclosed, the method comprising: and according to the information text generation model, arranging a key attribute information frame of the information article to generate an information text.
According to still another aspect of the present invention, there is disclosed an information text generation model construction apparatus adapted to construct an information text generation model using a neural network method, the apparatus comprising:
the map construction module is used for acquiring information article information, extracting key attribute information of the information article information and constructing an information knowledge map;
and the model generation module is used for carrying out graph convolution operation on the information knowledge graph according to the information knowledge graph to obtain an information text generation model.
According to yet another aspect of the present invention, there is provided a computing device comprising: one or more processors; and a memory; one or more programs, wherein the one or more programs are stored in memory and configured to be executed by one or more processors, the one or more programs comprising instructions for performing any of the methods of constructing an information text generation model as described above, and/or the one or more programs comprising instructions for performing any of the methods of generating information text using an information text generation model as described above.
According to yet another aspect of the present invention, there is provided a computer-readable storage medium storing one or more programs, the one or more programs comprising instructions, which when executed by a computing device, cause the computing device to perform any of the methods of constructing an information text generation model as described above, and/or the one or more programs comprising instructions for performing any of the methods of generating information text using an information text generation model as described above.
According to the construction scheme of the information text generation model, the key attribute information of the information article information is extracted by acquiring the information article information, and an information knowledge graph is constructed; and carrying out graph convolution operation on the information knowledge graph according to the information knowledge graph to obtain an information text generation model. The constructed information knowledge graph can be used as effective external knowledge to be integrated into the information graph in the subsequent information text generation process, graph convolution operation is carried out on the constructed knowledge graph, global knowledge representation of the knowledge graph is extracted, the knowledge representation is involved in decoding operation of a later decoder and is used as a hidden variable to guide a decoding calculation process, the purpose of controlling logical relations among sentences is achieved, meanwhile, the reasoning capacity of a graph neural network is exerted, the fact correctness of generated articles is guaranteed, the calculation result of the graph neural network is fused during decoding, decoding of sentence level and decoding of word level are completed, so that generated information content is logically clear in chapter structure, words in sentences are proper, and the expression content detail is rich.
Drawings
To the accomplishment of the foregoing and related ends, certain illustrative aspects are described herein in connection with the following description and the annexed drawings, which set forth the various ways in which the principles disclosed herein may be practiced, and all aspects and equivalents thereof are intended to fall within the scope of the claimed subject matter. The above, as well as additional objects, features, and advantages of the present disclosure will become more apparent from the following detailed description when read in conjunction with the accompanying drawings. Like reference numerals generally refer to like parts or elements throughout the present disclosure.
FIG. 1 illustrates a schematic construction of a computing device 100 according to one embodiment of the invention; and
FIG. 2 illustrates a flow diagram of a method 200 of constructing an information text generation model according to one embodiment of the invention; and
fig. 3 shows a schematic configuration of an information text generating apparatus 300 according to an embodiment of the present invention.
Detailed Description
Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
Fig. 1 is a block diagram of an example computing device 100. In a basic configuration 102, computing device 100 typically includes a system memory 106 and one or more processors 104. The memory bus 108 may be used for communication between the processor 104 and the system memory 106.
Depending on the desired configuration, the processor 104 may be any type of processing including, but not limited to: a microprocessor (μp), a microcontroller (μc), a digital information processor (DSP), or any combination thereof. The processor 104 may include one or more levels of caches, such as a first level cache 110 and a second level cache 112, a processor core 114, and registers 116. The example processor core 114 may include an Arithmetic Logic Unit (ALU), a Floating Point Unit (FPU), a digital signal processing core (DSP core), or any combination thereof. The example memory controller 118 may be used with the processor 104, or in some implementations, the memory controller 118 may be an internal part of the processor 104.
Depending on the desired configuration, system memory 106 may be any type of memory including, but not limited to: volatile memory (such as RAM), non-volatile memory (such as ROM, flash memory, etc.), or any combination thereof. The system memory 106 may include an operating system 120, one or more applications 122, and program data 124. In some implementations, the application 122 may be arranged to operate on an operating system with program data 124. In some embodiments, the computing device 100 is configured to execute the method 200 for constructing the information text generation model, where the method 200 can ensure that in the subsequent text generation process, the information text generation model can be integrated into the information text generation model as effective external knowledge, perform a graph convolution operation on the constructed knowledge graph, extract a global knowledge representation of the knowledge graph, participate in a decoding operation of a later decoder, guide a decoding calculation process as hidden variables, achieve the purpose of controlling a logical relationship between sentences, exert the inference capability of the graph neural network, ensure the fact correctness of generated articles, integrate the calculation result of the graph neural network during decoding, complete decoding at sentence level and decoding at word level, so that the generated information content has clear logic on chapter structure, has proper words in sentences, and has rich expression content details, and the program data 124 includes instructions for executing the method 200.
Computing device 100 may also include an interface bus 140 that facilitates communication from various interface devices (e.g., output devices 142, peripheral interfaces 144, and communication devices 146) to basic configuration 102 via bus/interface controller 130. The example output device 142 includes a graphics processing unit 148 and an audio processing unit 150. They may be configured to facilitate communication with various external devices such as a display or speakers via one or more a/V ports 152. Example peripheral interfaces 144 may include a serial interface controller 154 and a parallel interface controller 156, which may be configured to facilitate communication with external devices such as input devices (e.g., keyboard, mouse, pen, voice input device, touch input device) or other peripherals (e.g., printer, scanner, etc.) via one or more I/O ports 158. An example communication device 146 may include a network controller 160, which may be arranged to facilitate communication with one or more other computing devices 162 via one or more communication ports 164 over a network communication link.
The network communication link may be one example of a communication medium. Communication media may typically be embodied by computer readable instructions, data structures, program modules, and may include any information delivery media in a modulated data signal, such as a carrier wave or other transport mechanism. A "modulated data signal" may be a signal that has one or more of its data set or changed in such a manner as to encode information in the signal. By way of non-limiting example, communication media may include wired media such as a wired network or special purpose network, and wireless media such as acoustic, radio Frequency (RF), microwave, infrared (IR) or other wireless media. The term computer readable media as used herein may include both storage media and communication media. In some embodiments, one or more programs are stored in a computer readable medium, the one or more programs including instructions for performing methods by which the computing device 100 performs the method 200 of constructing an information text generation model, according to embodiments of the present invention.
Computing device 100 may be implemented as part of a small-sized portable (or mobile) electronic device such as a cellular telephone, a Personal Digital Assistant (PDA), a personal media player device, a wireless web-watch device, a personal headset device, an application specific device, or a hybrid device that may include any of the above functions. Computing device 100 may also be implemented as a personal computer including desktop and notebook computer configurations.
Text generation can be divided into open field text generation and vertical field text generation from the field; short text generation and long text generation can be classified in terms of the length of the generated text. The difficulty of generating vertical domain text and long text is relatively large, and the technical system involved is more complex.
Text generation is divided into two types from task categories, one is a data to text task based on a template or deep learning method. The goal of this task is to generate text from a structured data source, the subject and content of which are relatively well-defined, suitable for use in the vertical field. In the early implementation process, complex calculation is not generally involved, a large number of rules and templates are more applied, the robustness is high, and a developer needs to customize the templates in advance according to the requirements of users; with the development of deep learning, a method based on a deep learning model becomes a mainstream method of data to text, and the types of neural networks involved in the method are more, including but not limited to LSTM (Long Short-Term Memory) and variants thereof, variable self-encoders, attention mechanisms, graphic neural networks and the like. Another task is text generation based on a large-scale pre-training model, in the first step of implementation, the beginning part of the generated text is provided, and the generated content is divergent and is more suitable for the open field, and typical representatives are GPT2, GPT3, UNILM and the like.
The most mature method existing in the industry at present is based on a template text generation method, and the core idea of the method is to quickly generate a text segment by using organized structured data through a fixed template or rule. Represented by patent CN 101470700A, a text generation method is provided for formulating a corresponding template after determining the key slot position according to grammar rules. The scheme is widely applied in a plurality of fields, in the fields, the target text is often an objective fact of narration occurrence, and the text structure is relatively fixed, so that the method has extremely high running speed, can achieve the considerable speed of generating tens of thousands of grades per second, and the finally generated result is extremely reliable as long as the provided data are accurate; the method has the defects that the generated text is not abundant enough, the applicable field is limited, a large amount of manpower is required for designing templates, the maintenance work is heavy, the method is not suitable for being applied to scenes with large data volume, in the field with more subjects such as information, the generated articles of the templates are single in sentence pattern, the text plates are arranged, the user is finally caused to lose reading interest, and the method is unfavorable for media in the long term.
With the rapid development of deep learning, the text generation method based on the seq2seq architecture cannot meet the diversified demands of users, so that the text generation method based on the seq2seq is rapidly developed. The method is represented by the patent CN106980683A and the patent CN105930314A, a template-based generation method is abandoned, a coder-decoder-based text abstract generation method is constructed, in the simplest open field simple dialogue scene, only sufficient training data is needed, a coding end and a decoding end are usually simple RNN, GRU and other neural networks, a certain effect can be achieved, but in a text abstract and personalized dialogue system, the structure of the seq2seq is more complex, more external information is often added into a part of the coding end, themes or other external knowledge are fused, or details in the decoding process are controlled, and the change really makes the text generation result more satisfactory; the main defects of the scheme are that the text generation result is too simple to integrate with industry information, the complex scene of the field cannot be dealt with, and the problems of word repetition and unsmooth sentences are very easy to occur.
The text and training method based on the large-scale text and the transducer architecture also obtains good results along with the course of the transducer architecture in the field of natural language processing, the Chinese pre-training model based on the architecture is developed by google, hundred degrees, harbin industrial university and the like, and a user can obtain a complete section of more smooth text only by providing prefix words of target text; the main disadvantages of the scheme are that the pre-training Chinese model requires enough calculation power and has extremely high cost, and the obtained pre-training model cannot ensure the applicability of the pre-training model in the target field, and the generated text is smooth but often far away from the content to be expressed by the user.
FIG. 2 illustrates a flow chart of a method 200 of constructing an information text generation model according to one embodiment of the application. As shown in fig. 2, the method 200 is suitable for completing text generation of information by using a neural network, and the method 200 starts with step S210, obtaining information article information, extracting key attribute information of the information article information, and constructing an information knowledge graph.
Specifically, since the information is very extensive and the types of items are very numerous, after the information of the information article is obtained, the knowledge extraction algorithm is required to extract the key knowledge points in the information of the information article, and the specific knowledge extraction algorithm is one or more of the prior art, which is not specifically described herein. The information knowledge graph is constructed by collecting a large amount of information, including automobile news, information articles and the like, classifying according to class labels of the articles, removing useless information, such as forms, advertisements, editors, hyperlinks and the like, in the information of the information articles, extracting key attribute information of the processed information articles, and further constructing the information knowledge graph.
Specifically, according to an embodiment of the present invention, the step of obtaining information article information, extracting key attribute information of the information article information, and constructing an information knowledge graph includes:
acquiring information article information and classifying according to class labels of the information article information;
extracting key attribute information from the information article information to obtain the key attribute information of the information article information; specifically, the key attribute information is a key word in the information article information, and the main content of the information article information can be obtained through the key attribute information, and further, the core content in the information article information can be completely obtained through the expansion of the key word, and the secondary content is removed.
According to the key attribute information of the information article information, performing word segmentation and sentence segmentation operation on the information article information; specifically, the word segmentation and sentence segmentation operation aims at acquiring a keyword and a keyword sentence, thereby acquiring the core content of the information article information.
Acquiring a first clause set corresponding to the key attribute information according to the word segmentation and clause operation, wherein the first clause set comprises clauses corresponding to the key attribute information and the ordering of the key attribute information in the clauses; specifically, the first clause set is a corresponding sentence in the information article information obtained through the key attribute information of the information article information, and all the sentence sets corresponding to the key attribute information form the first clause set corresponding to the key attribute information. For example, the set of the key attribute information is set to be E, each element in E is a key attribute information, each key attribute information corresponds to a clause, and then the set of the clauses corresponding to all the key attribute information is set to be S, and then S is the first set of the clauses.
Acquiring a second clause set according to the first clause set, wherein the second clause set comprises global clauses formed by elements of all the first clause sets and key attribute information contexts corresponding to the clauses; specifically, after the first clause set is obtained, the required content in the information of the information article cannot be completely obtained, the clauses in the first clause set are only key sentences in the information of the information article, are not coherent but discrete, are not words of a word order, and the second clause set is used for combining the clauses in the first clause set with the context to form a word order coherent paragraph content, so that information knowledge conforming to Chinese grammar and the semantic meaning of the article is formed.
And constructing an information knowledge graph according to the first sentence set and the second sentence set. Specifically, the final information knowledge graph can be obtained through the clauses corresponding to the key attribute information in the first clause set and the key attribute information context corresponding to the clauses in the second clause set. The information knowledge graph at this time includes key attribute information, clauses corresponding to the key attribute information, and key attribute information context corresponding to the clauses corresponding to the key attribute information.
Specifically, in one embodiment of the present application, the step of extracting key attribute information of the information article information to obtain key attribute information of the information article information includes:
cleaning the information article information to remove useless information in the information article information; specifically, the garbage information comprises pictures, forms, comments, advertisements, author information, hyperlinks and the like in the information article information, and after the garbage information is removed, the cleaning of the information article information is completed. The classification and cleaning of the information articles are completed, and the data cleaning purpose is that the information articles are not interfered by junk texts in the process of constructing a later knowledge graph and training a model; the aim of text classification is to facilitate data analysis, find out unique key attribute information under each category, and design a starting data filling frame under each category.
Extracting keywords from the cleaned information article information to obtain a first key attribute set; specifically, the keywords of the information article information are generally determined according to the article semantics expressed by the information article information, and the keywords may be preset, for example, a certain structural component of the automobile, performance characteristics of the automobile, manufacturer of the automobile, brand of the automobile, etc. may be set as the keywords of the information article information, and when the keywords are extracted, the extracted keywords may be directly extracted from the information article information, or may be set by the user according to the content of the information article information. It should be noted that the keywords in the first set of key attributes are only discrete, non-sentence-forming words that are not semantically related to each other.
Performing main-predicate-guest triple extraction and causal triple extraction on the cleaned information article information, and performing de-duplication treatment on the extraction result to obtain a second key attribute set; specifically, by extracting the main predicate and the cause and effect triplet of the automobile, that is, obtaining the keywords more conforming to the semantic logic of the article, duplication may occur between the keywords extracted by the union and triplet and the keywords extracted by the cause and effect triplet, and thus, a deduplication process is required. It should be specifically noted that, a certain causal relationship or a main-guest relationship exists between one or more keywords in the second key attribute set, that is, the semantic logic or causal logic of the article is met.
Carrying out automobile domain entity identification on the cleaned information article information to obtain a third key attribute set; specifically, the entity identification in the automotive field is that the extracted keywords are keywords in the automotive field, that is, the entity identification in the information of the automotive article may be represented as an entity type and an entity value structure, for example, the third key attribute set may be represented as: { entity type-type, entity value-value } get structure.
And acquiring the key attribute information of the information article information according to the intersection of the elements in the first key attribute set, the second key attribute set and the third key attribute set. Specifically, by judging the intersection of the element in the third key attribute set with the first key attribute set and the second key attribute set, the required key attribute information can be finally obtained, and the representation form of the key attribute information is as follows: { entity type-type, entity value-value }.
And using a keyword extraction algorithm and a triplet extraction algorithm to comprehensively obtain important information in the information of the information article, extracting a key entity type-type and a corresponding entity value-value in the information of the information article by using an entity identification algorithm in the automotive field, and filtering out the information extracted in the keyword extraction algorithm and the triplet extraction algorithm respectively by using the key entity type-type and the corresponding entity value-value as a key attribute information set for constructing a knowledge graph and training data.
Specifically, in one embodiment of the present application, the step of extracting key attribute information from the information article information to obtain key attribute information of the information article information further includes:
and according to the classification of the information article information, compiling a regular expression to extract unique key attribute elements under the classification of the information article information, and taking the key attribute elements as supplements of the key attribute information. Specifically, since there may be some unique key attribute elements for a certain information article information, if the unique key attribute elements are not extracted, the integrity of text generation of the information article information will be affected, and therefore, after the extraction of the key attribute information of the information article information is completed, it is necessary to extract the unique key attribute elements under each category of each information article information by compiling a regular expression to supplement the key attribute information.
Specifically, in one embodiment of the present application, the step of obtaining a first clause set corresponding to the key attribute information according to the word segmentation and the clause operation, where the first clause set includes clauses corresponding to the key attribute information, and the step of ordering the key attribute information in the clauses includes:
performing word segmentation and sentence segmentation operation on the information article information, and extracting sentences corresponding to the key attribute information in the information article information; specifically, the word segmentation operation is to acquire a certain keyword in the key attribute information, and the sentence segmentation operation is to acquire a corresponding sentence through the keyword. By acquiring each keyword in the key attribute information and acquiring the clause corresponding to each keyword, a first clause set is obtained, and particularly, the clauses in the first clause set are discrete, incoherent and incompletely conform to Chinese semantics.
Sorting the key attribute information according to the sequence of occurrence of the key attribute information in the clause, and obtaining a fourth key attribute set; specifically, the key attribute information is ordered according to the sequence of the key attribute information in the clauses after the clauses obtained through the key attribute information are combined, and the fourth key attribute set is a set obtained by ordering the key attribute information in all the substations. The fourth key attribute set can be used for acquiring key attribute information combining information semantics and word order of the information articles, so that information texts can be acquired better, faster and more smoothly.
And acquiring a first clause set according to the clause and the fourth key attribute set. Specifically, a first clause set is formed according to all sentences corresponding to the key attribute information and the fourth key attribute set after the key attribute information in the sentences is sequenced, wherein the elements of the first clause set comprise all sentences corresponding to the key attribute information and the fourth key attribute set after the key attribute information in the sentences is sequenced.
The information text information is added into the knowledge graph, so that the key point is that one piece of information can be divided into several groups of sentences, the key attribute information contained in each group of sentences is included, and the sequence of the sentences and the key attribute information is also shown in the information knowledge graph.
Specifically, in one embodiment of the present application, the step of obtaining the second sentence set according to the first sentence set includes:
according to a first clause set, acquiring a global clause formed by all elements in the first clause set; specifically, all elements in the first clause set are built into a global clause, the global clause is a global semantic combined with information of the information article, and the global clause is formed after all elements in the first clause set are arranged.
Acquiring a corresponding key attribute information context of the global clause according to the global clause formed by all elements in the first clause set; specifically, the sentence combination with more accurate expression, more clean semantics and more reasonable word sequence can be obtained by acquiring the corresponding key attribute information context of the global clause.
And acquiring a second clause set according to the global clause and the corresponding key attribute information context of the global clause. Specifically, the second clause set includes a global clause and a context corresponding to the key attribute information in the global clause.
Specifically, by constructing the information knowledge graph, the knowledge extraction algorithm is used for extracting key knowledge points of the information article, so that the information knowledge graph can be used as effective external knowledge to be integrated into the generation process in the subsequent text generation process.
Through step S220, a graph convolution operation is performed on the information knowledge graph according to the information knowledge graph, so as to obtain an information text generation model. Specifically, after the information knowledge graph is obtained, the information knowledge graph is subjected to knowledge representation by using graph convolution operation, and then the information text generation model can be obtained by multi-layer neural network coding.
Specifically, in one embodiment of the present application, the step of performing a graph convolution operation on the information knowledge graph according to the information knowledge graph to obtain an information text generation model includes:
carrying out knowledge representation on the information knowledge graph by using a graph convolution network algorithm to obtain first representation global context information;
constructing a text generation network of a seq2seq framework, and coding each piece of key attribute information by using a bidirectional GRU to obtain a first coding set; specifically, the seq2seq framework is a generic encoder and decoder framework for machine translation, text summarization, session modeling, image description, and the like, the GRU (Gate Recurrent Unit, gate cyclic array) is one of RNNs (Recurrent Neural Network, cyclic neural network), the RNN is a type of recursive neural network that uses sequence data as input, recursions in the evolution direction of the sequence, and all nodes are chained, and the first encoding set is a set obtained by encoding each key attribute information using bidirectional GRUs. The text generation network for constructing the seq2seq frame brings global context information of the information knowledge graph to the decoder, and the calculated result of the auxiliary decoder is more consistent with the knowledge and logic expressed by the information knowledge graph.
Connecting the first coding set by using a multi-layer neural network, and calculating to obtain a first plan decoder; specifically, the first coding set is connected through the multi-layer neural network to operate, so that a plan decoder can be obtained, and the plan decoder at the moment is the first plan decoder.
According to the first representation global context information and the first plan decoder, obtaining a first operation output through GRU operation, wherein an operation result at any moment in the first operation output is related to an operation process at the next moment; specifically, for example, a first representation global context information is defined as Z p The first plan decoder is represented asThen by putting Z p And->The first operation output can be obtained by GRU operation, and is expressed as g t Let the first operation output at a certain time t be g t G is then t G involved in time t+1 t+1 Is calculated by the computer.
Acquiring a first hidden variable at any momentThe first hidden variable at any moment is obtained by the first operation output, the first representation global context information and the first hidden variable at the previous moment through a variational self-encoder; specifically, it is assumed that the first hidden variable at a certain time t is expressed as And->The first hidden variable +_for a certain time t corresponds to the Gaussian distribution>From a first operation output g t First representation global context information Z p First hidden variable +.>Obtained from the encoder by variation. In the decoder, a sentence-level decoder and a word-level decoder are constructed, so that the overlevel from coarse granularity to fine granularity is ensured when the text is generated, the rationality of text structure logic is ensured, and the smoothness of each paragraph or sentence is ensured.
According to the first representation global context information, performing sentence-level decoding operation on the first hidden variable at any moment by using the GRU, taking a decoding result as calculation input of the first hidden variable at the next moment, and participating in decoding;
and performing iterative training until the loss value converges, and obtaining an information text generation model, wherein the generated text index of the information text generation model is larger than a set threshold value.
It should be understood that, although the steps in the flowchart of fig. 2 are shown in sequence as indicated by the arrows, the steps are not necessarily performed in sequence as indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps in fig. 2 may include multiple sub-steps or stages that are not necessarily performed at the same time, but may be performed at different times, nor do the order in which the sub-steps or stages are performed necessarily performed in sequence, but may be performed alternately or alternately with at least a portion of the sub-steps or stages of other steps or other steps.
In accordance with another aspect of the present application, there is also disclosed a method of generating an information text using an information text generation model, the method comprising: and according to the information text generation model, arranging a key attribute information frame of the information article to generate an information text.
Specifically, by using the information text generation model, the information text can be effectively generated by using the information text, in a specific execution process, a key attribute information frame under the corresponding category of the article is arranged by taking a main body of the generated target text as a center, the key attribute information is perfected, and a perfected key attribute information set is input into the trained model, so that the required information text information is generated.
Filling key attribute information as starting data, constructing a key attribute information frame according to information article information categories and attribute pairs specific to each category, for example, under the category of 'new vehicle marketing', the attribute is marked as a specific attribute when the information of other categories does not exist in the category of 'marketing time' and 'marketing place'; the attributes "train", "brand" and "equation" are attribute information that is almost always covered under each category, such attributes are labeled as generic attributes. The specific attribute and the general attribute form key attribute information together, the key attribute information is edited by a user and then input into a trained model to generate a result, so that the generated result statement is smooth, and errors which are easily corrected, such as word repetition, are avoided.
In one embodiment, as shown in fig. 3, there is provided a construction apparatus 300 of an information text generation model, the apparatus 300 including: and the map construction module and the model generation module.
The map construction module is used for acquiring information article information, extracting key attribute information of the information article information and constructing an information knowledge map; and the model generation module is used for carrying out graph convolution operation on the information knowledge graph according to the information knowledge graph to obtain an information text generation model.
Specifically, in another embodiment of the present application, the map construction module is configured to obtain information articles and classify the information articles according to category labels of the information articles; extracting key attribute information from the information article information to obtain the key attribute information of the information article information; according to the key attribute information of the information article information, performing word segmentation and sentence segmentation operation on the information article information; acquiring a first clause set corresponding to the key attribute information according to the word segmentation and clause operation, wherein the first clause set comprises clauses corresponding to the key attribute information and the ordering of the key attribute information in the clauses; acquiring a second clause set according to the first clause set, wherein the second clause set comprises global clauses formed by elements of all the first clause sets and key attribute information contexts corresponding to the clauses; and constructing an information knowledge graph according to the first sentence set and the second sentence set.
Specifically, in another embodiment of the present application, the map construction module is configured to clean the information article information to remove useless information in the information article information; extracting keywords from the cleaned information article information to obtain a first key attribute set; performing main-predicate-guest triple extraction and causal triple extraction on the cleaned information article information, and performing de-duplication treatment on the extraction result to obtain a second key attribute set; carrying out automobile domain entity identification on the cleaned information article information to obtain a third key attribute set; and acquiring the key attribute information of the information article information according to the intersection of the elements in the first key attribute set, the second key attribute set and the third key attribute set.
Specifically, in another embodiment of the present application, the map construction module is configured to write a regular expression according to the classification of the information article information to extract key attribute elements unique to each classification of the information article information, as a supplement to the key attribute information.
Specifically, in another embodiment of the present application, the map construction module is configured to perform word segmentation and sentence segmentation on the information article information, and extract a sentence corresponding to the key attribute information in the information article information; sorting the key attribute information according to the sequence of occurrence of the key attribute information in the clause, and obtaining a fourth key attribute set; and acquiring a first clause set according to the clause and the fourth key attribute set.
Specifically, in another embodiment of the present application, the map construction module is configured to obtain, according to a first clause set, a global clause formed by all elements in the first clause set; acquiring a corresponding key attribute information context of the global clause according to the global clause formed by all elements in the first clause set; and acquiring a second clause set according to the global clause and the corresponding key attribute information context of the global clause.
Specifically, in another embodiment of the present application, the model generating module is configured to use a graph convolution network algorithm to perform knowledge representation on the information knowledge graph to obtain first representation global context information; constructing a text generation network of a seq2seq framework, and coding each piece of key attribute information by using a bidirectional GRU to obtain a first coding set; connecting the first coding set by using a multi-layer neural network, and calculating to obtain a first plan decoder; according to the first representation global context information and the first plan decoder, obtaining a first operation output through GRU operation, wherein an operation result at any moment in the first operation output is related to an operation process at the next moment; acquiring a first hidden variable at any moment, wherein the first hidden variable at any moment is obtained by the first operation output, the first representation global context information and a first hidden variable at the previous moment through a variational self-encoder; according to the first representation global context information, performing sentence-level decoding operation on the first hidden variable at any moment by using the GRU, taking a decoding result as calculation input of the first hidden variable at the next moment, and participating in decoding; and performing iterative training until the loss value converges, and obtaining an information text generation model, wherein the generated text index of the information text generation model is larger than a set threshold value.
A7 the method of A1, wherein the step of organizing the key attribute information frames of the information articles according to the information text generation model, and generating the information text comprises:
carrying out knowledge representation on the information knowledge graph by using a graph convolution network algorithm to obtain first representation global context information;
constructing a text generation network, and coding each piece of key attribute information to obtain a first coding set;
connecting the first coding set by using a multi-layer neural network, and calculating to obtain a first plan decoder;
acquiring a first operation output according to the first representation global context information and the first plan decoder, wherein an operation result at any moment in the first operation output is related to an operation process at the next moment;
acquiring a first hidden variable at any moment, wherein the first hidden variable at any moment is obtained by the first operation output, the first representation global context information and a first hidden variable at the previous moment through a variational self-encoder;
according to the first representation global context information, performing sentence-level decoding operation on the first hidden variable at any moment, taking a decoding result as calculation input of the first hidden variable at the next moment, and participating in decoding;
And performing iterative training until the loss value converges, and obtaining an information text generation model, wherein the generated text index of the information text generation model is larger than a set threshold value.
It should be appreciated that in the above description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof for the purpose of streamlining the disclosure and aiding in the understanding of one or more of the various inventive aspects. However, the disclosed method should not be construed as reflecting the intention that: i.e., the claimed invention requires more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive aspects lie in less than all features of a single foregoing disclosed embodiment. Thus, the claims following the detailed description are hereby expressly incorporated into this detailed description, with each claim standing on its own as a separate embodiment of this invention.
Those skilled in the art will appreciate that the modules or units or components of the devices in the examples disclosed herein may be arranged in a device as described in this embodiment, or alternatively may be located in one or more devices different from the devices in this example. The modules in the foregoing examples may be combined into one module or may be further divided into a plurality of sub-modules.
Those skilled in the art will appreciate that the modules in the apparatus of the embodiments may be adaptively changed and disposed in one or more apparatuses different from the embodiments. The modules or units or components of the embodiments may be combined into one module or unit or component and, furthermore, they may be divided into a plurality of sub-modules or sub-units or sub-components. Any combination of all features disclosed in this specification (including any accompanying claims, abstract and drawings), and all of the processes or units of any method or apparatus so disclosed, may be used in combination, except insofar as at least some of such features and/or processes or units are mutually exclusive. Each feature disclosed in this specification (including any accompanying claims, abstract and drawings), may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise.
Furthermore, those skilled in the art will appreciate that while some embodiments described herein include some features but not others included in other embodiments, combinations of features of different embodiments are meant to be within the scope of the invention and form different embodiments. For example, in the following claims, any of the claimed embodiments can be used in any combination.
Furthermore, some of the embodiments are described herein as methods or combinations of method elements that may be implemented by a processor of a computer system or by other means of performing the functions. Thus, a processor with the necessary instructions for implementing the described method or method element forms a means for implementing the method or method element. Furthermore, the elements of the apparatus embodiments described herein are examples of the following apparatus: the apparatus is for carrying out the functions performed by the elements for carrying out the objects of the invention.
As used herein, unless otherwise specified the use of the ordinal terms "first," "second," "third," etc., to describe a general object merely denote different instances of like objects, and are not intended to imply that the objects so described must have a given order, either temporally, spatially, in ranking, or in any other manner.
While the invention has been described with respect to a limited number of embodiments, those skilled in the art, having benefit of the above description, will appreciate that other embodiments are contemplated within the scope of the invention as described herein. Furthermore, it should be noted that the language used in the specification has been principally selected for readability and instructional purposes, and may not have been selected to delineate or circumscribe the inventive subject matter. Accordingly, many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the appended claims. The disclosure of the present invention is intended to be illustrative, but not limiting, of the scope of the invention, which is defined by the appended claims.
Claims (9)
1. A method of constructing an information text generation model, the method being adapted to construct the information text generation model using a neural network method, the method comprising the steps of:
acquiring information article information, extracting key attribute information of the information article information, and constructing an information knowledge graph;
performing graph convolution operation on the information knowledge graph according to the information knowledge graph to obtain an information text generation model;
the step of obtaining information article information, extracting key attribute information of the information article information, and constructing an information knowledge graph, comprising the following steps:
acquiring information article information and classifying according to class labels of the information article information;
extracting key attribute information from the information article information to obtain the key attribute information of the information article information;
according to the key attribute information of the information article information, performing word segmentation and sentence segmentation operation on the information article information;
acquiring a first clause set corresponding to the key attribute information according to the word segmentation and clause operation, wherein the first clause set comprises clauses corresponding to the key attribute information and the ordering of the key attribute information in the clauses;
Acquiring a second clause set according to the first clause set, wherein the second clause set comprises global clauses formed by elements of all the first clause sets and key attribute information contexts corresponding to the clauses;
constructing an information knowledge graph according to the first sentence set and the second sentence set;
the method comprises the steps of carrying out graph convolution operation on the information knowledge graph according to the information knowledge graph to obtain an information text generation model, and comprises the following steps:
carrying out knowledge representation on the information knowledge graph by using a graph convolution network algorithm to obtain first representation global context information;
constructing a text generation network, and coding each piece of key attribute information to obtain a first coding set;
connecting the first coding set by using a multi-layer neural network, and calculating to obtain a first plan decoder;
acquiring a first operation output according to the first representation global context information and the first plan decoder, wherein an operation result at any moment in the first operation output is related to an operation process at the next moment;
acquiring a first hidden variable at any moment, wherein the first hidden variable at any moment is obtained by the first operation output, the first representation global context information and a first hidden variable at the previous moment through a variational self-encoder;
According to the first representation global context information, performing sentence-level decoding operation on the first hidden variable at any moment, taking a decoding result as calculation input of the first hidden variable at the next moment, and participating in decoding;
and performing iterative training until the loss value converges, and obtaining an information text generation model, wherein the generated text index of the information text generation model is larger than a set threshold value.
2. The method of claim 1, wherein the step of extracting key attribute information of the information article information to obtain key attribute information of the information article information comprises:
cleaning the information article information to remove useless information in the information article information;
extracting keywords from the cleaned information article information to obtain a first key attribute set;
performing main-predicate-guest triple extraction and causal triple extraction on the cleaned information article information, and performing de-duplication treatment on the extraction result to obtain a second key attribute set;
carrying out automobile domain entity identification on the cleaned information article information to obtain a third key attribute set;
and acquiring the key attribute information of the information article information according to the intersection of the elements in the first key attribute set, the second key attribute set and the third key attribute set.
3. The method of claim 2, wherein the step of extracting key attribute information from the information article information to obtain key attribute information of the information article information further comprises:
and according to the classification of the information article information, compiling a regular expression to extract unique key attribute elements under the classification of the information article information, and taking the key attribute elements as supplements of the key attribute information.
4. The method as claimed in claim 1, wherein the step of obtaining a first clause set corresponding to the key attribute information according to the word segmentation and sentence segmentation operations, the first clause set including the clauses corresponding to the key attribute information, and the ordering of the key attribute information in the clauses includes:
performing word segmentation and sentence segmentation operation on the information article information, and extracting sentences corresponding to the key attribute information in the information article information;
sorting the key attribute information according to the sequence of occurrence of the key attribute information in the clause, and obtaining a fourth key attribute set;
and acquiring a first clause set according to the clause and the fourth key attribute set.
5. The method of claim 1, wherein the step of obtaining a second set of clauses from the first set of clauses comprises:
According to a first clause set, acquiring a global clause formed by all elements in the first clause set;
acquiring a corresponding key attribute information context of the global clause according to the global clause formed by all elements in the first clause set;
and acquiring a second clause set according to the global clause and the corresponding key attribute information context of the global clause.
6. A method of generating an information text using an information text generation model, the method comprising: an information text generation model generated according to the information text generation model construction method of any one of claims 1 to 5, collating key attribute information frames of information articles, and generating an information text.
7. An apparatus for constructing an information text generation model, the apparatus being adapted to construct the information text generation model using a neural network method, the apparatus comprising:
the map construction module is used for acquiring information article information, extracting key attribute information of the information article information and constructing an information knowledge map;
the model generation module is used for carrying out graph convolution operation on the information knowledge graph according to the information knowledge graph to obtain an information text generation model;
The step of obtaining information article information, extracting key attribute information of the information article information, and constructing an information knowledge graph, comprising the following steps:
acquiring information article information and classifying according to class labels of the information article information;
extracting key attribute information from the information article information to obtain the key attribute information of the information article information;
according to the key attribute information of the information article information, performing word segmentation and sentence segmentation operation on the information article information;
acquiring a first clause set corresponding to the key attribute information according to the word segmentation and clause operation, wherein the first clause set comprises clauses corresponding to the key attribute information and the ordering of the key attribute information in the clauses;
acquiring a second clause set according to the first clause set, wherein the second clause set comprises global clauses formed by elements of all the first clause sets and key attribute information contexts corresponding to the clauses;
constructing an information knowledge graph according to the first sentence set and the second sentence set;
the method comprises the steps of carrying out graph convolution operation on the information knowledge graph according to the information knowledge graph to obtain an information text generation model, and comprises the following steps:
Carrying out knowledge representation on the information knowledge graph by using a graph convolution network algorithm to obtain first representation global context information;
constructing a text generation network, and coding each piece of key attribute information to obtain a first coding set;
connecting the first coding set by using a multi-layer neural network, and calculating to obtain a first plan decoder;
acquiring a first operation output according to the first representation global context information and the first plan decoder, wherein an operation result at any moment in the first operation output is related to an operation process at the next moment;
acquiring a first hidden variable at any moment, wherein the first hidden variable at any moment is obtained by the first operation output, the first representation global context information and a first hidden variable at the previous moment through a variational self-encoder;
according to the first representation global context information, performing sentence-level decoding operation on the first hidden variable at any moment, taking a decoding result as calculation input of the first hidden variable at the next moment, and participating in decoding;
and performing iterative training until the loss value converges, and obtaining an information text generation model, wherein the generated text index of the information text generation model is larger than a set threshold value.
8. An electronic device, comprising:
one or more processors; and
a memory;
one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs comprising instructions for performing any of the methods of claims 1-6.
9. A computer readable storage medium storing one or more programs, the one or more programs comprising instructions, which when executed by a computing device, cause the computing device to perform any of the methods of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110955326.XA CN113626614B (en) | 2021-08-19 | 2021-08-19 | Method, device, equipment and storage medium for constructing information text generation model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110955326.XA CN113626614B (en) | 2021-08-19 | 2021-08-19 | Method, device, equipment and storage medium for constructing information text generation model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113626614A CN113626614A (en) | 2021-11-09 |
| CN113626614B true CN113626614B (en) | 2023-10-20 |
Family
ID=78386716
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110955326.XA Active CN113626614B (en) | 2021-08-19 | 2021-08-19 | Method, device, equipment and storage medium for constructing information text generation model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113626614B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114997143B (en) * | 2022-08-04 | 2022-11-15 | 北京澜舟科技有限公司 | Text generation model training method and system, text generation method and storage medium |
| CN117992601A (en) * | 2024-04-03 | 2024-05-07 | 紫金诚征信有限公司 | Document generation method and device based on artificial intelligence |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109960749A (en) * | 2019-02-22 | 2019-07-02 | 清华大学 | Model acquisition methods, keyword generation method, device, medium and calculating equipment |
| CN110287278A (en) * | 2019-06-20 | 2019-09-27 | 北京百度网讯科技有限公司 | Comment on generation method, device, server and storage medium |
| CN110825869A (en) * | 2019-09-16 | 2020-02-21 | 中山大学 | Text abstract generating method of variation generation decoder based on copying mechanism |
| CN111930929A (en) * | 2020-07-09 | 2020-11-13 | 车智互联（北京）科技有限公司 | Article title generation method and device and computing equipment |
| CN112084383A (en) * | 2020-09-07 | 2020-12-15 | 中国平安财产保险股份有限公司 | Information recommendation method, device and equipment based on knowledge graph and storage medium |
| CN112567355A (en) * | 2018-09-04 | 2021-03-26 | 北京京东尚科信息技术有限公司 | End-to-end structure-aware convolutional network for knowledge base completion |
| CN112949302A (en) * | 2021-02-25 | 2021-06-11 | 平安国际智慧城市科技股份有限公司 | Long text generation method, device, terminal and storage medium |
-
2021
- 2021-08-19 CN CN202110955326.XA patent/CN113626614B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112567355A (en) * | 2018-09-04 | 2021-03-26 | 北京京东尚科信息技术有限公司 | End-to-end structure-aware convolutional network for knowledge base completion |
| CN109960749A (en) * | 2019-02-22 | 2019-07-02 | 清华大学 | Model acquisition methods, keyword generation method, device, medium and calculating equipment |
| CN110287278A (en) * | 2019-06-20 | 2019-09-27 | 北京百度网讯科技有限公司 | Comment on generation method, device, server and storage medium |
| CN110825869A (en) * | 2019-09-16 | 2020-02-21 | 中山大学 | Text abstract generating method of variation generation decoder based on copying mechanism |
| CN111930929A (en) * | 2020-07-09 | 2020-11-13 | 车智互联（北京）科技有限公司 | Article title generation method and device and computing equipment |
| CN112084383A (en) * | 2020-09-07 | 2020-12-15 | 中国平安财产保险股份有限公司 | Information recommendation method, device and equipment based on knowledge graph and storage medium |
| CN112949302A (en) * | 2021-02-25 | 2021-06-11 | 平安国际智慧城市科技股份有限公司 | Long text generation method, device, terminal and storage medium |
Non-Patent Citations (3)
| Title |
|---|
| SK-GCN: Modeling Syntax and Knowledge via Graph Convolutional Network for aspect-level sentiment classification;Jie Zhou et al.;《Knowledge-Based Systems》;1-10 * |
| 基于序列生成模型的语法纠错技术研究;王辰成;《中国优秀硕士学位论文全文数据库 哲学与人文科学辑》;F084-11 * |
| 基于深度学习的短时交通流预测研究;《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》;《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》;C034-675 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113626614A (en) | 2021-11-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108280061B (en) | Text processing method and device based on ambiguous entity words | |
| CN107133213B (en) | Method and system for automatically extracting text abstract based on algorithm | |
| CN107463607B (en) | Method for acquiring and organizing upper and lower relations of domain entities by combining word vectors and bootstrap learning | |
| CN111680094B (en) | Text structuring method, device and system and non-volatile storage medium | |
| CN108806671B (en) | Semantic analysis, device and electronic equipment | |
| CN113626614B (en) | Method, device, equipment and storage medium for constructing information text generation model | |
| CN111538828B (en) | Text emotion analysis method, text emotion analysis device, computer device, and readable storage medium | |
| CN115048944B (en) | Open domain dialogue reply method and system based on theme enhancement | |
| CN111091000A (en) | Processing system and method for extracting user fine-grained typical opinion data | |
| CN113626713A (en) | Search method, device, equipment and storage medium | |
| CN112101004A (en) | General webpage character information extraction method based on conditional random field and syntactic analysis | |
| CN111339772B (en) | Russian text emotion analysis method, electronic device and storage medium | |
| CN116521825A (en) | Method for generating text matching model, computing device and storage medium | |
| CN116108840A (en) | Text fine granularity emotion analysis method, system, medium and computing device | |
| Zhao et al. | POS-ATAEPE-BiLSTM: an aspect-based sentiment analysis algorithm considering part-of-speech embedding | |
| CN115906824A (en) | Text fine-grained emotion analysis method, system, medium and computing equipment | |
| CN114969341A (en) | Fine-grained emotion analysis method and device for catering industry comments | |
| CN112182159A (en) | Personalized retrieval type conversation method and system based on semantic representation | |
| CN105808522A (en) | Method and apparatus for semantic association | |
| CN110929513A (en) | Text-based label system construction method and device | |
| CN110610001A (en) | Short text integrity identification method and device, storage medium and computer equipment | |
| Tang et al. | Short text classification combining keywords and knowledge | |
| Wu et al. | Cm-tcn: channel-aware multi-scale temporal convolutional networks for speech emotion recognition | |
| CN116842168B (en) | Cross-domain problem processing method and device, electronic equipment and storage medium | |
| CN117436457B (en) | Irony identification method, irony identification 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 |