CN116304098A - Intelligent generation matching theme and information method based on commodity - Google Patents
Intelligent generation matching theme and information method based on commodity Download PDFInfo
- Publication number
- CN116304098A CN116304098A CN202310275371.XA CN202310275371A CN116304098A CN 116304098 A CN116304098 A CN 116304098A CN 202310275371 A CN202310275371 A CN 202310275371A CN 116304098 A CN116304098 A CN 116304098A
- Authority
- CN
- China
- Prior art keywords
- commodity
- data
- information
- text
- soft
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 54
- 238000012549 training Methods 0.000 claims description 13
- 238000005070 sampling Methods 0.000 claims description 8
- 230000008451 emotion Effects 0.000 claims description 7
- 230000006870 function Effects 0.000 claims description 7
- 230000009193 crawling Effects 0.000 claims description 6
- 238000009792 diffusion process Methods 0.000 claims description 6
- 238000005065 mining Methods 0.000 claims description 5
- MJIHNNLFOKEZEW-UHFFFAOYSA-N lansoprazole Chemical compound CC1=C(OCC(F)(F)F)C=CN=C1CS(=O)C1=NC2=CC=CC=C2N1 MJIHNNLFOKEZEW-UHFFFAOYSA-N 0.000 claims description 4
- 238000004140 cleaning Methods 0.000 claims description 3
- 238000013508 migration Methods 0.000 claims description 3
- 230000005012 migration Effects 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 4
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000007405 data analysis Methods 0.000 description 2
- 238000007418 data mining Methods 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 230000032683 aging Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
Images
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
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation
-
- 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
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
- G06F16/353—Clustering; Classification into predefined classes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0276—Advertisement creation
-
- 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
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Business, Economics & Management (AREA)
- Software Systems (AREA)
- Development Economics (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Business, Economics & Management (AREA)
- Human Computer Interaction (AREA)
- Game Theory and Decision Science (AREA)
- General Health & Medical Sciences (AREA)
- Economics (AREA)
- Marketing (AREA)
- Entrepreneurship & Innovation (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Health & Medical Sciences (AREA)
- Animal Behavior & Ethology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a method for intelligently generating matched topics and information based on commodities, and relates to the technical field of computers. In order to solve the problems of how to keep news generation fresh and making a search engine feel original in the prior art; a method for intelligently generating matched subjects and information based on commodities comprises the following steps: step one: inputting a theme; step two: outputting a basic article; step three: outputting multiple articles; by combining knowledge graph forms with the information of the existing commodity portraits, corresponding commodity libraries are expanded for each platform to respectively form commodity multi-mode knowledge graphs of each platform, information of the same product on different platforms is obtained, a generation model driven by the knowledge base is trained, the multi-mode knowledge graphs of the commodity of the knowledge base are combined, pictures and texts in details are input into the model at the same time, news of different styles are generated, and therefore the effects of keeping fresh and enabling a search engine to feel original are achieved.
Description
Technical Field
The invention relates to the technical field of computers, in particular to an intelligent generation matching theme and information method based on commodities.
Background
In the operation process of each platform, operators need to find pictures and contents related to commodities to write and release the contents every day, and the work is repeated and the effect is not high; regarding a method for intelligently generating matching information, related patents, such as chinese patent publication No. CN115298660a, disclose an artificial intelligence-based information matching method, apparatus, device, medium and program product, wherein the method comprises: acquiring information content; determining an information topic label of the information content based on a preset algorithm model; judging whether a target theme label matched with the information theme label exists in the first gallery or not, wherein the picture in the first gallery is marked with the corresponding theme label; if the first gallery has the target theme label matched with the information theme label, taking the picture corresponding to the target theme label as a cover map of the information content.
The above patent, while improving the efficiency and effectiveness of information mapping, still has the following problems:
in the prior art, the difficulty of how to keep news generation fresh and enable a search engine to feel original often exists, and when a generated model is trained, the simplest training mode is to send marked data into a language model for fine adjustment, but the method is high in cost, and the operation cost of each platform is greatly improved.
Disclosure of Invention
The invention aims to provide a method for intelligently generating matched topics and information based on commodities so as to solve the problems in the background technology.
In order to achieve the above purpose, the present invention provides the following technical solutions:
a method for intelligently generating matched topics and information based on commodities, the method comprising the following steps:
step one: theme input: determining an input theme, determining commodities based on the theme, matching the commodities with an existing product library and commodity images, inputting detailed data corresponding to the commodities into the existing product library and commodity images, matching the detailed data one by one, and drawing a knowledge graph;
acquiring commodity libraries corresponding to all the platforms, respectively forming commodity multi-mode knowledge maps of all the platforms, aligning the commodities in the commodity multi-mode knowledge maps, and establishing a generation model driven by the knowledge libraries;
step two: basic article output: crawling corresponding commodity soft texts from the acquired multi-mode knowledge patterns, and learning a writing method of the commodity soft texts based on the generated model, wherein the number of the commodity soft texts is at least two;
performing migration learning on the trained generative model by adopting a generative pre-training task, generating according to the generative model driven by a knowledge base, training a reward model, classifying the model into abcd grades, scoring the generated soft text, and manually selecting the optimal abcd by human intervention at the moment;
step three: multi-article output: and searching by combining a search engine, comparing the difference degree between generated news to serve as a loss function, extracting corresponding line text style characteristics and emotion characteristics in the commodity soft text through learning of a model layer, and generating news of different styles through prompt learning of Promp.
Further, based on the generated model driven by the knowledge base, when obtaining one commodity detail, combining the multi-mode knowledge graph of the commodity in the knowledge base, and simultaneously inputting the picture and the text in the commodity detail into the generated model.
Further, in the first step, the matching of the commodity with the existing product library and commodity portraits is specifically as follows:
acquiring actively input theme data from each platform terminal, acquiring a data identifier of the theme data, and inputting the data identifier of the theme data into a preset theme database for matching;
outputting basic data types of commodity data of the theme based on the matching result, wherein the preset theme database comprises basic data types of commodity names, commodity pictures and commodity text descriptions;
preprocessing commodity data based on basic data types of the commodity data to obtain corresponding matching results contained in the commodity data, wherein the matching results are not less than one;
establishing a corresponding commodity storage space based on the matching result, binding commodity detailed data of a commodity pre-stored detailed information database with commodities, and respectively acquiring pre-stored detailed commodity information corresponding to each commodity;
inputting the pre-stored detailed commodity information into a corresponding commodity storage space, establishing commodity information, and simultaneously, based on data associated parameters carried by commodities in the commodity storage space;
and extracting detailed commodity data of each commodity in the corresponding commodity storage space, combining corresponding data association parameters, acquiring association relations among different commodity storage spaces, and drawing a knowledge graph corresponding to the commodity based on the association relations.
Further, aiming at the commodity multi-mode knowledge graph formed in the first step, the method comprises the following steps:
acquiring commodity databases corresponding to all platforms, inputting data in the commodity storage space into the commodity databases, and classifying and storing based on basic data types of the data in the commodity storage space;
according to the commodity database corresponding to each platform, after the commodity database is stored, the associated information of the commodity is called, wherein the associated information comprises: similar commodity picture information, similar commodity description information and associated data characteristic information;
and generating a knowledge graph of each commodity according to the data of each commodity in the commodity database and the commodity corresponding to the association relation, and filling the data of the corresponding commodity into the knowledge graph of each commodity based on the association relation to obtain a commodity multi-mode knowledge graph.
Further, in the multi-mode knowledge graph of the commodity, when the data of any commodity is called, detailed information of the corresponding commodity is called according to the multi-mode knowledge graph of the commodity, and meanwhile, the related commodity under the related information of the commodity is independently presented in the form of the knowledge graph.
Further, for learning the writing method of the commodity soft text based on the generated model in the second step, the writing method specifically comprises the following steps:
crawling soft text data and parameter data in a period of time, and determining a writing weight value of each soft text according to node attribute and parameter data of each soft text data and a generated pre-training task;
creating a filtered word stock, and performing word segmentation on the text and the title of the soft text data to obtain a plurality of extracted words, and determining the word characteristics of each extracted word and the sentence characteristics of each extracted word in the text and the title;
determining the similarity between the extracted words based on the constructed synonym dictionary, the word characteristics of each extracted word and the sentence characteristics of each extracted word in the text and the title, and screening the words based on the similarity;
and cleaning words according to the part-of-speech statistical features to obtain a soft text main body, transversely mining soft text data based on the soft text main body, calculating writing features in each soft text by using writing weights, and determining a plurality of writing feature parameters based on the writing features.
Further, aiming at the search combined with the search engine in the third step, the difference degree between the generated news is compared as a loss function, and the specific process comprises the following steps:
text generation: from language model P lm Sampling in (W), wherein w= [ W ] 1 ...w n ]Is a series of discrete words, P 1m (W) is a distribution about word sequences;
controllable text generation: sampling from a conditional distribution p (w|c), where c represents a control attribute.
Further, the controllable text generation further includes: and establishing a diffusion model, and generating controllable text attributes based on the diffusion model.
Compared with the prior art, the invention has the beneficial effects that:
1. the method comprises the steps of combining information of existing commodity portraits in a knowledge graph mode, expanding corresponding commodity libraries for each platform to form commodity multi-mode knowledge graphs of each platform, conducting entity alignment on commodities in the knowledge graphs to obtain information of the same product on different platforms, training a generation model driven by the knowledge bases, combining the multi-mode knowledge graphs of the commodities in the knowledge bases, inputting pictures and texts in details into the model at the same time, learning line style features and emotion features of news through learning of model layers, prompting learning through Promp, and generating news of different styles, so that the effects of keeping freshness and enabling a search engine to feel original are achieved.
2. The basic data type of commodity data of a theme is determined, the commodity data is preprocessed, the commodity detailed data of a commodity pre-stored detailed information database is utilized to bind with the commodity, the detailed information and the corresponding picture information of the commodity can be known, and the association characteristics of the commodity and similar commodities are determined by combining the corresponding data association parameters, so that the commodity and the related commodity are displayed more comprehensively, the basis is made for data analysis of the commodity and soft writing of the commodity, and the comprehensiveness of commodity data is improved.
3. The commodity display has a unified format through the knowledge graph, the commodity multi-mode knowledge graph can be formed only by filling the knowledge graph, the management of detailed data of the commodity is effectively improved, the simplicity of the commodity data in the knowledge graph is improved through displaying any commodity data in the knowledge graph, the sensitivity to main data is improved, people can obtain effective information in the commodity multi-mode knowledge graph more easily, and in addition, the whole and part of conversion is realized by retrieving the data of any commodity, so that the knowledge graph is convenient to display and observe.
4. By calculating the weight value of the soft text writing method for multiple times, different soft text writing method information is accurately mined and learned from mass data, the problems that part of information is easy to filter and the mining range is not comprehensive in the traditional data mining method can be solved, a large number of soft text templates are preset after the soft text writing method is learned, soft text is automatically generated based on the structural information of commodities, the problems that the traditional soft text editing is slow in timeliness and low in originality are solved, and the freshness and originality of the generated soft text are kept.
Drawings
FIG. 1 is a flow chart of a method for intelligently generating matching topics and information based on commodities according to the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
In order to solve the technical problem that in the prior art, how to keep news generation fresh and make a search engine feel original is often existed, referring to fig. 1, the present embodiment provides the following technical scheme:
a method for intelligently generating matched topics and information based on commodities, the method comprising the following steps:
step one: theme input: determining an input theme, determining commodities based on the theme, matching the commodities with an existing product library and commodity images, inputting detailed data corresponding to the commodities into the existing product library and commodity images, matching the detailed data one by one, and drawing a knowledge graph; acquiring commodity libraries corresponding to all the platforms, respectively forming commodity multi-mode knowledge maps of all the platforms, aligning the commodities in the commodity multi-mode knowledge maps, and establishing a generation model driven by the knowledge libraries; based on a generated model driven by a knowledge base, when obtaining a commodity detail, combining a multi-mode knowledge graph of the commodity in the knowledge base, and simultaneously inputting pictures and texts in the commodity detail into the generated model;
step two: basic article output: crawling corresponding commodity soft texts from the acquired multi-mode knowledge patterns, and learning a writing method of the commodity soft texts based on the generated model, wherein the number of the commodity soft texts is at least two; performing migration learning on the trained generative model by adopting a generative pre-training task, generating according to the generative model driven by a knowledge base, training a reward model, classifying the model into abcd grades, scoring the generated soft text, and manually selecting the optimal abcd by human intervention at the moment;
step three: multi-article output: and searching by combining a search engine, comparing the difference degree between generated news to serve as a loss function, extracting corresponding line text style characteristics and emotion characteristics in the commodity soft text through learning of a model layer, and generating news of different styles through prompt learning of Promp.
Specifically, the method comprises the steps of adding commodity details into the existing product library and commodity images through the existing product library, combining information of the existing commodity images through a knowledge graph mode, expanding corresponding commodity libraries for each platform, respectively forming commodity multi-mode knowledge graphs of each platform, conducting entity alignment on commodities in the knowledge graphs to obtain information of the same product on different platforms, training a generation model driven by the knowledge base, combining the multi-mode knowledge graphs of the commodities in the knowledge library, inputting pictures and texts in the details into the model at the same time, learning the character and emotion characteristics of news through learning of the model layer, and generating news of different styles through prompt learning, so that the effects of keeping freshness and enabling a search engine to feel original are achieved.
In order to solve the technical problem that in the prior art, since a large amount of commodity information and data need to be acquired, other commodities related to the commodity cannot be judged when the system extracts the commodity information, and writing of corresponding soft texts is affected, referring to fig. 1, the present embodiment provides the following technical scheme:
aiming at matching the commodity with the existing product library and commodity portraits in the first step, the method specifically comprises the following steps:
acquiring actively input theme data from each platform terminal, acquiring a data identifier of the theme data, and inputting the data identifier of the theme data into a preset theme database for matching; outputting basic data types of commodity data of the theme based on the matching result, wherein the preset theme database comprises basic data types of commodity names, commodity pictures and commodity text descriptions;
preprocessing commodity data based on basic data types of the commodity data to obtain corresponding matching results contained in the commodity data, wherein the matching results are not less than one; establishing a corresponding commodity storage space based on the matching result, binding commodity detailed data of a commodity pre-stored detailed information database with commodities, and respectively acquiring pre-stored detailed commodity information corresponding to each commodity;
inputting the pre-stored detailed commodity information into a corresponding commodity storage space, establishing commodity information, and simultaneously, based on data associated parameters carried by commodities in the commodity storage space; and extracting detailed commodity data of each commodity in the corresponding commodity storage space, combining corresponding data association parameters, acquiring association relations among different commodity storage spaces, and drawing a knowledge graph corresponding to the commodity based on the association relations.
Specifically, the basic data type of commodity data of a theme is determined, the commodity data is preprocessed, commodity detailed data of a commodity pre-stored detailed information database is utilized to bind with the commodity, detailed information and corresponding picture information of the commodity can be known, and the association characteristics of the commodity and similar commodities are determined by combining corresponding data association parameters, so that the commodity and the related commodities are displayed more comprehensively, the data analysis of the commodity is carried out subsequently, the basis is based on soft text writing of the commodity, and the comprehensiveness of the commodity data is improved.
In order to solve the technical problem that the structure is huge and the display of commodity center data cannot be performed when the knowledge graph displays a large amount of commodities and detailed data related to the commodities in the prior art, referring to fig. 1, the present embodiment provides the following technical scheme:
aiming at the commodity multi-mode knowledge graph which is formed in the first step and comprises the following steps:
acquiring commodity databases corresponding to all platforms, inputting data in the commodity storage space into the commodity databases, and classifying and storing based on basic data types of the data in the commodity storage space; according to the commodity database corresponding to each platform, after the commodity database is stored, the associated information of the commodity is called, wherein the associated information comprises: similar commodity picture information, similar commodity description information and associated data characteristic information; generating a knowledge graph of each commodity according to the data of each commodity in the commodity database and the commodity corresponding to the association relation, and filling the data of the corresponding commodity into the knowledge graph of each commodity based on the association relation to obtain a multi-mode knowledge graph of the commodity; in the commodity multi-mode knowledge graph, when the data of any commodity is called, the detailed information of the corresponding commodity is called according to the commodity multi-mode knowledge graph, and the related commodity under the commodity related information is independently presented in the form of the knowledge graph.
Specifically, the commodity display can be provided with a uniform format through the knowledge graph, the commodity multi-mode knowledge graph can be formed only by filling the knowledge graph, the management of detailed data of the commodity is effectively improved, the simplicity of the commodity data in the knowledge graph is improved through displaying any commodity data in the knowledge graph, the sensitivity of the main data is improved, people can obtain effective information in the commodity multi-mode knowledge graph more easily, and in addition, the whole and part of conversion is realized through retrieving the data of any commodity, so that the knowledge graph is convenient to present and observe.
In order to solve the technical problems of slow editing aging and low originality in the prior art, please refer to fig. 1, the present embodiment provides the following technical scheme:
aiming at the writing method for learning the commodity soft text based on the generated model in the second step, the method specifically comprises the following steps:
crawling soft text data and parameter data in a period of time, and determining a writing weight value of each soft text according to node attribute and parameter data of each soft text data and a generated pre-training task; creating a filtered word stock, and performing word segmentation on the text and the title of the soft text data to obtain a plurality of extracted words, and determining the word characteristics of each extracted word and the sentence characteristics of each extracted word in the text and the title;
determining the similarity between the extracted words based on the constructed synonym dictionary, the word characteristics of each extracted word and the sentence characteristics of each extracted word in the text and the title, and screening the words based on the similarity; and cleaning words according to the part-of-speech statistical features to obtain a soft text main body, transversely mining soft text data based on the soft text main body, calculating writing features in each soft text by using writing weights, and determining a plurality of writing feature parameters based on the writing features.
Specifically, by extracting keywords from mass data and calculating the weight value of the soft text writing method for multiple times, different soft text writing method information is accurately mined and learned from the mass data, the problem that part of information is easy to filter and the mining range is incomplete in a traditional data mining method can be solved, a large number of soft text templates are preset after the soft text writing method is learned, soft text is automatically generated based on structural information of commodities, the problems that the traditional soft text editing time is slow and the originality is low are solved, and the freshness and originality of the generated soft text are kept.
In order to solve the technical problem that in the prior art, the simplest training method is to send the markup data into the language model for fine tuning, but the cost of the method is expensive, referring to fig. 1, the present embodiment provides the following technical scheme:
aiming at the search combined with a search engine in the third step, comparing the difference degree between generated news as a loss function, wherein the specific process comprises the following steps:
text generation: from language model P 1m Sampling in (W), wherein w= [ W ] 1 ...w n ]Is a series of discrete words, P lm (W) is a distribution about word sequences; controllable text generation: sampling from a conditional distribution p (w|c), where c represents the control property, namely: syntax parsing tree in syntax control or expected emotion in emotion control; the controllable text generation further includes: establishing a diffusion model, and generating controllable text attributes based on the diffusion model;
in this embodiment, a plug-and-play controllable generation: p (w|c) ≡P lm (W)*p(c|w)。
Specifically, the text is generated by sampling from the language model, the controllable text is generated by sampling from the condition distribution, the attribute control is completed while the fluency is effectively ensured, meanwhile, the plug-and-play controllable generation method is derived, the output of the preview model is controlled, the traditional method of sending the mark data into the language model for fine tuning is avoided, the cost is greatly reduced, the search engine can be quickly combined for searching, and the difference degree between generated news is compared as a loss function.
The foregoing is only a preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art, who is within the scope of the present invention, should be covered by the protection scope of the present invention by making equivalents and modifications to the technical solution and the inventive concept thereof.
Claims (8)
1. A method for intelligently generating matched subjects and information based on commodities is characterized in that: the method comprises the following steps:
step one: theme input: determining an input theme, determining commodities based on the theme, matching the commodities with an existing product library and commodity images, inputting detailed data corresponding to the commodities into the existing product library and commodity images, matching the detailed data one by one, and drawing a knowledge graph;
acquiring commodity libraries corresponding to all the platforms, respectively forming commodity multi-mode knowledge maps of all the platforms, aligning the commodities in the commodity multi-mode knowledge maps, and establishing a generation model driven by the knowledge libraries;
step two: basic article output: crawling corresponding commodity soft texts from the acquired multi-mode knowledge patterns, and learning a writing method of the commodity soft texts based on the generated model, wherein the number of the commodity soft texts is at least two;
performing migration learning on the trained generative model by adopting a generative pre-training task, generating according to the generative model driven by a knowledge base, training a reward model, classifying the model into abcd grades, scoring the generated soft text, and manually selecting the optimal abcd by human intervention at the moment;
step three: multi-article output: and searching by combining a search engine, comparing the difference degree between generated news to serve as a loss function, extracting corresponding line text style characteristics and emotion characteristics in the commodity soft text through learning of a model layer, and generating news of different styles through prompt learning of Promp.
2. The intelligent commodity-based method for generating matching topics and information according to claim 1, wherein: and based on the generated model driven by the knowledge base, when one commodity detail is obtained, combining the multi-mode knowledge graph of the commodity in the knowledge base, and simultaneously inputting the picture and the text in the commodity detail into the generated model.
3. The intelligent commodity-based method for generating matching topics and information as claimed in claim 2, wherein: aiming at matching the commodity with the existing product library and commodity portraits in the first step, the method specifically comprises the following steps:
acquiring actively input theme data from each platform terminal, acquiring a data identifier of the theme data, and inputting the data identifier of the theme data into a preset theme database for matching;
outputting basic data types of commodity data of the theme based on the matching result, wherein the preset theme database comprises basic data types of commodity names, commodity pictures and commodity text descriptions;
preprocessing commodity data based on basic data types of the commodity data to obtain corresponding matching results contained in the commodity data, wherein the matching results are not less than one;
establishing a corresponding commodity storage space based on the matching result, binding commodity detailed data of a commodity pre-stored detailed information database with commodities, and respectively acquiring pre-stored detailed commodity information corresponding to each commodity;
inputting the pre-stored detailed commodity information into a corresponding commodity storage space, establishing commodity information, and simultaneously, based on data associated parameters carried by commodities in the commodity storage space;
and extracting detailed commodity data of each commodity in the corresponding commodity storage space, combining corresponding data association parameters, acquiring association relations among different commodity storage spaces, and drawing a knowledge graph corresponding to the commodity based on the association relations.
4. A method for intelligently generating matching topics and information based on commodities as in claim 3, wherein: aiming at the commodity multi-mode knowledge graph which is formed in the first step and comprises the following steps:
acquiring commodity databases corresponding to all platforms, inputting data in the commodity storage space into the commodity databases, and classifying and storing based on basic data types of the data in the commodity storage space;
according to the commodity database corresponding to each platform, after the commodity database is stored, the associated information of the commodity is called, wherein the associated information comprises: similar commodity picture information, similar commodity description information and associated data characteristic information;
and generating a knowledge graph of each commodity according to the data of each commodity in the commodity database and the commodity corresponding to the association relation, and filling the data of the corresponding commodity into the knowledge graph of each commodity based on the association relation to obtain a commodity multi-mode knowledge graph.
5. The intelligent commodity-based method for generating matching topics and information as claimed in claim 4, wherein: in the commodity multi-mode knowledge graph, when the data of any commodity is called, the detailed information of the corresponding commodity is called according to the commodity multi-mode knowledge graph, and the related commodity under the commodity related information is independently presented in the form of the knowledge graph.
6. The intelligent commodity-based method for generating matching topics and information as claimed in claim 5, wherein: aiming at the writing method for learning the commodity soft text based on the generated model in the second step, the method specifically comprises the following steps:
crawling soft text data and parameter data in a period of time, and determining a writing weight value of each soft text according to node attribute and parameter data of each soft text data and a generated pre-training task;
creating a filtered word stock, and performing word segmentation on the text and the title of the soft text data to obtain a plurality of extracted words, and determining the word characteristics of each extracted word and the sentence characteristics of each extracted word in the text and the title;
determining the similarity between the extracted words based on the constructed synonym dictionary, the word characteristics of each extracted word and the sentence characteristics of each extracted word in the text and the title, and screening the words based on the similarity;
and cleaning words according to the part-of-speech statistical features to obtain a soft text main body, transversely mining soft text data based on the soft text main body, calculating writing features in each soft text by using writing weights, and determining a plurality of writing feature parameters based on the writing features.
7. The intelligent commodity-based method for generating matching topics and information as claimed in claim 6, wherein: aiming at the search combined with a search engine in the third step, comparing the difference degree between generated news as a loss function, wherein the specific process comprises the following steps:
text generation: from language model P lm Sampling in (W), wherein w= [ W ] 1 ...w n ]Is a series of discrete words, P lm (W) is a distribution about word sequences;
controllable text generation: sampling from a conditional distribution p (w|c), where c represents a control attribute.
8. The intelligent commodity-based method for generating matching topics and information as claimed in claim 7, wherein: the controllable text generation further includes: and establishing a diffusion model, and generating controllable text attributes based on the diffusion model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310275371.XA CN116304098A (en) | 2023-03-15 | 2023-03-15 | Intelligent generation matching theme and information method based on commodity |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310275371.XA CN116304098A (en) | 2023-03-15 | 2023-03-15 | Intelligent generation matching theme and information method based on commodity |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116304098A true CN116304098A (en) | 2023-06-23 |
Family
ID=86793930
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310275371.XA Pending CN116304098A (en) | 2023-03-15 | 2023-03-15 | Intelligent generation matching theme and information method based on commodity |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116304098A (en) |
-
2023
- 2023-03-15 CN CN202310275371.XA patent/CN116304098A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11947621B2 (en) | System and method for the creation and update of hierarchical websites based on collected business knowledge | |
| CN103377258B (en) | Method and apparatus for carrying out classification display to micro-blog information | |
| KR20220006491A (en) | Method, apparatus, electronic device, storage medium and computer program for generating comment subtitle | |
| CN111553138B (en) | Auxiliary writing method and device for standardizing content structure document | |
| CN114048335B (en) | A user interaction method and device based on knowledge base | |
| CN112163560A (en) | Video information processing method and device, electronic equipment and storage medium | |
| CN117436414A (en) | Presentation generation method and device, electronic equipment and storage medium | |
| CN117746143A (en) | AIGC-based image description text generation method and device and storage medium | |
| CN104881428B (en) | A kind of hum pattern extraction, search method and the device of hum pattern webpage | |
| CN114840657A (en) | API knowledge graph self-adaptive construction and intelligent question-answering method based on mixed mode | |
| CN114818639A (en) | Presentation generation method, device, device and storage medium | |
| Hollink | Semantic annotation for retrieval of visual resources | |
| CN119149762A (en) | Data processing method, device, electronic equipment and storage medium | |
| JP7549706B2 (en) | DATA EXPANSION SYSTEM, DATA EXPANSION METHOD, AND PROGRAM | |
| CN111626023A (en) | Automatic generation method, device and system for visualization chart highlighting and annotation | |
| CN116304098A (en) | Intelligent generation matching theme and information method based on commodity | |
| CN116362331A (en) | Knowledge point filling method based on man-machine cooperation construction knowledge graph | |
| CN113821651A (en) | Information processing method and device | |
| CN114911898B (en) | Search method, device and electronic device based on knowledge graph | |
| US12299028B2 (en) | Systems and methods for identifying a design template matching a search query | |
| CN118760763B (en) | Method and device for generating accompanying reading manuscripts based on extraction of knowledge points at different grades | |
| EP4571565A1 (en) | Systems and methods for processing designs | |
| JP2001184356A (en) | User taste learning type data base retrieving device | |
| CN120067381A (en) | Method and device for generating electronic picture book, storage medium, and computer equipment | |
| CN119850789A (en) | Image material generation method, device, medium and computing equipment |
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 |