CN117171191A

CN117171191A - Intelligent search engine system and method in material vertical field

Info

Publication number: CN117171191A
Application number: CN202311027082.4A
Authority: CN
Inventors: 张博明; 赵晓杰; 孙洁; 王言; 颛孙学堃; 刘诚; 刘辉
Original assignee: Shandong Industry Research Institute Of Advanced Materials Co ltd
Current assignee: Shandong Industry Research Institute Of Advanced Materials Co ltd
Priority date: 2023-08-15
Filing date: 2023-08-15
Publication date: 2023-12-05

Abstract

The invention provides an intelligent search engine system and method in the vertical field of materials, and relates to the field of intelligent search. The system comprises a graph calculation search analysis module, a graph calculation search analysis module and a graph calculation search analysis module, wherein the graph calculation search analysis module is used for acquiring a user search intention based on analysis indexes selected by a user, acquiring user question-answer intentions based on questions input by the user, understanding the user search intention and the user question-answer intentions, and carrying out real-time clustering, correlation analysis and attribution analysis to obtain data analysis results and predicted answers; the multi-round question-answering module sends the user questions to the graph calculation, search and analysis module, receives the predicted answers and pushes the predicted answers to the user; the cognitive search module acquires analysis indexes selected by a user, sends the analysis indexes to the graph calculation search analysis module, receives data analysis results fed back by the graph calculation search analysis module, and presents the data analysis results to the user. The invention solves the problems that data in the material field cannot be shared and mass data are explored and analyzed through knowledge graph and search engine technology, and can analyze more intelligently and solve the internal demands of personnel in the material field.

Description

Intelligent search engine system and method in material vertical field

Technical Field

The invention belongs to the technical field of intelligent search, and particularly relates to an intelligent search engine system and method in the vertical field of materials.

Background

The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.

Materials, energy sources and information are three major supports of modern science and technology. The material has wide application, is permeated into various industries, such as industrial articles, civil articles, household appliances, electronic products, network communication, pharmaceutical chemical industry, equipment manufacturing, aerospace equipment, sanitation, environmental protection and the like, has no important relation with the material, has close relation with the preparation, properties, application and the like of the material in many fields, and is a material foundation for technological development and human society progress.

The material field includes the technical principles of production activities and internal operations caused by the elements such as institutions, personnel and products related to materials. With the development of the material field, a great deal of data, information and knowledge is accumulated. The circulation and exchange of such data, information and knowledge is important to innovative advances in the materials arts. The life cycle of the material industry data is embodied in various links such as data generation, collection, processing, transmission, searching, application and the like. In the internet age today, more and more users use search engines to retrieve material data, making data delivery and application more efficient. The vertical field search engine greatly shortens the data propagation distance, reduces the propagation cost, enlarges the propagation total quantity and forms forward feedback with the integral application benefit. In the field of material verticality, users using search engines generally have a professional background of materiality, they need both unstructured text information and precise structured data, and even most of users focus on the latest leading edge audience data, so the retrieval requirements of the user population are more difficult to meet.

The inventor finds that at present, when industrial data in the domestic material field is searched and analyzed, no better platform support exists, and the data cannot be shared, so that the operation is very difficult when institutions such as upstream and downstream of each industry, research institutions and the like find related data of the material industry and the like.

In addition, when related searches in the material field are performed by using other universal search engines, a search result in a link form is often obtained, unstructured data is obtained, and a searcher needs to perform secondary searches based on the search result in the link form, so that the operation is inconvenient. And the arrangement of the search results is affected by many factors, and it takes time for the searcher to find the search contents desired by himself. Meanwhile, because the universal search engine lacks basic knowledge support of the material field, the material field cannot be well matched for asking when man-machine multi-round interaction asking and answering is performed, the overall search efficiency is low, intelligent analysis is not possible, the internal requirements of personnel in the material field are solved, the search results presented for the searcher are not satisfactory, and the method cannot be directly applied to reports in the material field.

Disclosure of Invention

In order to overcome the defects of the prior art, the invention provides an intelligent search engine system and method in the vertical field of materials, which solves the problems that data in the material field cannot be shared and mass data are explored and analyzed through knowledge graph and search engine technology, and can analyze more intelligently and solve the internal demands of personnel in the material field.

To achieve the above object, one or more embodiments of the present invention provide the following technical solutions:

the first aspect of the invention provides an intelligent search engine system in the vertical field of materials.

A material verticality domain intelligent search engine system, comprising:

the multi-round question-answering module is used for sending the interactive behaviors and the use scenes of the user to the algorithm recommendation module and the graph calculation, search and analysis module, receiving the problems fed back by the graph calculation, search and analysis module and pushing the problems to the user; meanwhile, the questions input by the user after selection are sent to an algorithm recommendation module and a graph calculation, search and analysis module, and the predicted answers fed back by the graph calculation, search and analysis module are received and pushed to the user to complete multi-round dialogue with the user;

the algorithm recommendation module is used for receiving the user interaction behavior and the use scene sent by the multi-round question-answering module and the problems input by the user after selection, analyzing and matching the user interaction behavior and the use scene with the content of the material field in the database, carrying out correlation dimension calculation and association relation filtering of the graph structure, sequencing the algorithms in the database according to the correlation and the association degree of the user interaction behavior and the use scene and the input problems after selection by the user from high to low, and recommending the algorithm with the front sequencing as a first algorithm to the graph calculation search analysis module;

The graph calculation search analysis module is used for receiving a first algorithm recommended by the algorithm recommendation module, acquiring a problem with high matching degree with a user based on the interaction behavior and the use scene of the first algorithm and the user, pushing the problem to the multi-round question-answering module, acquiring the question-answering intention of the user based on the first algorithm and the problem input by the user, understanding the question-answering intention of the user, carrying out real-time clustering, correlation analysis and attribution analysis to obtain a predicted answer, and sending the predicted answer to the multi-round question-answering module.

Preferably, the system further comprises a cognitive search module, wherein:

the cognitive search module is used for configuring analysis indexes in a self-defined mode, acquiring analysis indexes selected by a user, sending the analysis indexes to the algorithm recommendation module and the graph calculation search analysis module respectively, receiving data analysis results fed back by the graph calculation search analysis module, and presenting the data analysis results to the user;

the algorithm recommendation module is used for receiving the analysis index selected by the user and sent by the cognitive search module, analyzing analysis index data selected by the user, calculating, filtering and sorting from high to low algorithms in a database through the graph engine, and sending the algorithm with the front sorting as a second algorithm to the graph calculation, search and analysis module;

The diagram calculation search analysis module is used for receiving a second algorithm recommended by the algorithm recommendation module, acquiring a user search intention based on analysis indexes selected by a user, carrying out real-time clustering, correlation analysis and attribution analysis on the user search intention to obtain multi-dimensional data analysis results including content results, knowledge card results and chart analysis results, and sending the data analysis results to the cognitive search module; the content results include informational web pages, patents, journals, reports, pictures, videos, and material analysis reports.

Preferably, the interaction behavior of the user comprises searching, clicking, staying and collecting behaviors, and the use scene comprises information, pictures, patents, journals and reported search pages.

Preferably, the graph calculation, search and analysis module is further configured to set the questions with high matching degree with the user in a multi-level form and set weights of each level according to multiple influencing factors, and when pushing to the multi-round question and answer module, the questions are pushed in priority according to the weights of the multi-level questions, where the multiple influencing factors include whether there is data, commercial value of the data, and hot spot of the material.

Preferably, the graph calculation search analysis module is further configured to generate a material analysis report meeting the user requirement by analyzing the user search intention, matching the current material name and industry in the material field database with the user search intention, and acquiring the template text and related field data from the material field database.

Preferably, the graph calculation search analysis module is further used for responding to the question of the user for the attribute, generating the association relationship between the edge and the concept according to the attribute relationship based on the graph calculation method, and forming the knowledge card results of the reasoning, the people and the institutions of the product.

Preferably, the graph analysis results support various graphical presentations including tables, bar graphs, pie charts.

Preferably, the graph calculation search analysis module is used for understanding the user search intention and the user question-answer intention based on a graph calculation method, and carrying out real-time clustering, correlation analysis and attribution analysis to obtain a data analysis result and a prediction answer.

Preferably, the system further comprises a user portrayal module, a map management module, an intelligent report module, a knowledge management module, a data docking module and a data application module, wherein:

the user portrait module is used for collecting data of basic information of a user in a user registration stage, and collecting data of searching, clicking, staying and collecting behaviors of the user in a user use stage to form a user portrait and a user label;

The map management module is used for carrying out visual management, map display, map version management, map design and map modification on the map;

the intelligent report module is used for managing a report template, generating a report and downloading the report;

the knowledge management module is used for realizing knowledge classification management, knowledge template management and knowledge flow management;

the data docking module is used for providing a data docking tool and realizing docking with webpage data, picture data, data and product data;

and the data application module is used for providing automatic generation of data insight reports, application market display and ordering functions.

The second aspect of the invention provides a method for an intelligent search engine in the vertical field of materials.

A material vertical field intelligent search engine method comprises the following steps:

the multi-round question-answering module sends the interactive behaviors and the use scenes of the user to the algorithm recommendation module and the graph calculation, search and analysis module;

the algorithm recommendation module receives user interaction behaviors and usage scenes, analyzes and matches the content of the material field in the database, performs correlation dimension calculation and association relation filtering of the graph structure, sorts the algorithms in the database according to the correlation and association degree between the user interaction behaviors and the usage scenes from high to low, and recommends the algorithm with the front sorting as a first algorithm to the graph calculation, search and analysis module;

The graph calculation, search and analysis module acquires a problem with high matching degree with a user based on a first algorithm, interaction behaviors of the user and a use scene and pushes the problem to the multi-round question-answering module;

the multi-round question-answering module pushes the questions with high matching degree with the user to the user, and simultaneously sends the questions input by the user after selection to the algorithm recommendation module and the graph calculation search analysis module;

the algorithm recommending module receives the problem input by the user after selection and recommends the first algorithm to the graph calculation, search and analysis module again;

the graph calculation search analysis module acquires user question and answer intentions based on a first algorithm and questions input by a user, understands the user question and answer intentions, performs real-time clustering, correlation analysis and attribution analysis to obtain predicted answers, and sends the predicted answers to the multi-round question and answer module;

the multi-round question-answering module receives the predicted answers and pushes the predicted answers to the user, and multi-round conversations with the user are completed.

The one or more of the above technical solutions have the following beneficial effects:

the invention relates to a method for analyzing and presenting data around the content of billions of material fields, which comprises multiple data dimensions such as products, raw materials, test items, institutions, personnel, industries, information, patents, reports, journals and the like, and establishes a structured database belonging to a material vertical field system as a bottom data source of the material vertical engine.

The invention provides an intelligent search engine system and a method in the vertical field of materials, wherein the system comprises user portraits, multi-round questions and answers, an algorithm recommendation module, intelligent recommendation, cognitive search, a data application center, intelligent report, map management, knowledge management and a map calculation search analysis module, and the function technology modules further contain functions with finer granularity, so that the combination of data analysis and intelligent search is realized, data products meeting the requirements of users are not disconnected, the product functions in the vertical field of materials are better realized, the blank of the search engine in the vertical field of materials at present is filled, and the search work of scientific researchers and technicians is greatly facilitated.

The invention uses multi-round machine question-answering and user interaction to intelligently recommend the required data of product data, industry trend prediction, customer demand prediction, literature summary report, industry research report and the like for the user, simultaneously satisfies the requirements of exploratory analysis on the data, supports the custom configuration of analysis indexes, can intelligently understand the search intention of the user, positions the multi-dimensional data analysis results conforming to the question, supports the graphic display of tables, bar charts, pie charts and the like, can conveniently obtain the search results without secondary search, and simultaneously displays the search results more conforming to the industry habit in the material field, thereby being convenient to operate and saving time and labor.

According to the invention, the multi-round question-answering module sends the interactive behaviors and the use scenes of the user to the graph calculation, search and analysis module, receives the problems fed back by the graph calculation, search and analysis module and pushes the problems to the user, so that a scientific and complete problem list can be provided for the user, the situation that the user searches for the field which is not known well, and how to put out the search problems is not clear is avoided, and the intelligent method is realized.

The graph calculation search analysis module is used for obtaining template text and related field data from the material field database by analyzing the user search intention and matching the current material name and industry in the material field database, so as to generate a material analysis report, solve the problem that the filling of the analysis report is time-consuming and labor-consuming by manual operation at present, and greatly facilitate the user.

Additional aspects of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.

Drawings

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention.

Fig. 1 is a system configuration diagram of a first embodiment.

Fig. 2 is a general structural diagram of the first embodiment.

Fig. 3 is a technical construction diagram of the first embodiment.

Fig. 4 is a deployment architecture diagram of the first embodiment.

Detailed Description

It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the invention. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the present invention.

Embodiments of the invention and features of the embodiments may be combined with each other without conflict.

The invention provides a general idea:

the invention designs an intelligent search engine system and method for the vertical field of materials through knowledge graph and search engine technology, wherein the system comprises user portraits, multiple questions and answers, intelligent recommendation, cognitive search, a data application center, intelligent report, graph management, knowledge management, graph calculation search analysis module and algorithm recommendation module, and the accuracy, comprehensiveness, professionality and individualization of the content in the current field are improved from different dimensions.

By establishing the operation engine platform in the vertical field of the material, the problem that data and contents in the material field cannot be well fused is solved, the problems that the data cannot be shared and massive data are explored and analyzed are solved, the searching accuracy and coverage of the material contents are improved, the data are deeply analyzed, and a data report is generated by aggregation, so that academic and research support is provided for the material field, and the internal requirements of personnel in the material field can be more intelligently analyzed and solved.

The invention is based on the database system built in the vertical field of materials and the structured data stored in the corresponding database, and comprises massive and comprehensive data of the material industry, and the details refer to the Chinese patent application with the patent number of CN202211102043.1 and the name of 'a database system for comprehensively describing the material industry'.

Example 1

The embodiment discloses an intelligent search engine system in the vertical field of materials.

As shown in fig. 1, a material vertical domain intelligent search engine system includes:

Further, the system also comprises a cognitive search module, wherein:

Graph computation is an abstract representation of a real world "graph" structure based on "graph theory" and the computational model on such data structures. In general, in graph computation, the basic data structure expression is: g= (V, E); v=vertex (vertex or node), e=edge. The graph data structure well expresses the relevance between data, so that problems in many applications can be abstracted into graphs to be represented, and the problems are solved by using ideas of graph theory or building a model based on the graphs.

The graph calculation solves the problems of real-time calculation, storage and management of graph data and the like. The traditional relational data expose the problems of modeling defects, horizontal expansion and contraction and the like, so that the complex relational computation is supported by adopting graph computation with more powerful expression capability.

In this embodiment, the method further includes a database, technical problems related to a plurality of material fields are prestored in the database, after the graph calculation, search and analysis module receives interaction behaviors and use scenes of the user, the graph calculation, search and analysis module firstly analyzes and understands what the target problem of the user is, and then invokes the problem prestored in the database and having high matching degree with the user to push the problem to the multi-round question-answer module.

Meanwhile, answers of a plurality of technical questions related to the material field are prestored in the database, after the graph calculation, search and analysis module receives questions input by a user, the user questions and answers are understood first, answers in the database corresponding to the user questions and answers are called based on the user questions and answers, and after the answers are returned through a knowledge graph and a search interface according to preset questions and answers and according to user selection, the answers are pushed to the multi-round questions and answers module.

Further, the interactive behaviors of the user comprise searching, clicking, staying and collecting behaviors, and the use scene comprises information, pictures, patents, journals and reported searching pages.

Further, the graph calculation, search and analysis module is further configured to set the questions in a multi-level form and set weights of each level according to multiple influencing factors, and when the questions with high matching degree with the actual searching of the user are pushed to the multi-round question and answer module, the questions are pushed in priority according to the weights of the multi-level questions, where the multiple influencing factors include whether data exist, commercial values of the data, and hot spots of the materials.

Further, the content results include informational web pages, patents, journals, reports, pictures, videos, reports, and materials analysis reports; the graph calculation search analysis module is also used for generating a material analysis report by analyzing the user search word and matching the current material name and industry in the material field database with the user search word and acquiring the template text and related field data from the material field database.

Further, the graph calculation search analysis module is further used for responding to the question of the user on the attribute, generating the association relation between the edge and the concept according to the attribute relation, splitting the data into the attribute and the concept, classifying the association relation by splitting, and forming query judgment on the data after the association relation is made, so that the query judgment is used for knowledge cards of people and all institutions of products.

Further, the graph analysis results support various graph displays including tables, bar graphs and pie charts.

Further, the graph calculation search analysis module is used for understanding the user search word analysis and the user question and answer based on the graph calculation method, and carrying out real-time clustering, correlation analysis and attribution analysis to obtain a data analysis result and an answer.

Further, still include user portrait module, map management module, intelligent report module, knowledge management module, data butt joint module and data application module, wherein:

Furthermore, the multi-round question-answering module is also used for acquiring the type of the user's work unit and the department to which the user belongs, and sending the acquired type of the user's work unit and the part to the user portrait module to enrich portrait information.

Specific:

1) Multiple rounds of questions and answers are matched through single search word results, accurate questions and answers, automatic replies, greetings, multiple rounds of automatic questions and the like. The weight of the multi-level problem can be set according to various influencing factors, such as whether data exists, the commercial value of the data, the hot spot of the material and the like, and the priority pushing of the problem can be carried out by referring to the weight.

The multi-round question and answer design in the product is divided into a PC end and a mobile phone applet end, and the multi-round question and answer can assist the material staff to search the content related to the material by carrying the material search engine, and meanwhile, the corresponding result can be pushed according to the content required by the material staff. For example, the user behavior recognition of the material person can be performed, the question and answer welcome and the question in the material field are automatically triggered, and meanwhile, the portrait information of the user is enriched.

The multi-round question-and-answer integrated question can surround the content of information, patents, journals, reports and the like in the material field, and the material properties with finer granularity, such as inorganic matters, organic matters, metals, glass and the like, so that a user can select a question to be understood, and meanwhile, the problem that the user in the material field does not know how to search the question is solved.

The multiple questions and answers can intelligently select the questions to automatically ask according to the interactive behaviors and the use scenes of the users. When the content direction is clearly inquired, such as searching pages of information, pictures, patents, journals, reports and the like, a plurality of questions and answers are clicked to inquire according to the scene page where the questions are located, such as when the information pages are searched, whether the information web page content is wanted to be known or not is inquired by the plurality of questions and answers, and meanwhile material properties possibly wanted to be known are given, so that the psychological and demand of the user are more met.

The multi-round question and answer can enrich portrait information according to user behaviors and scene pushing problems, can enable users to fill in and select own work unit types and affiliated departments, such as government units, material enterprises, universities and colleges, industry associations, parts and investment institutions and other units, and affiliated departments, such as purchasing, technology, sales, production, maneuver, enterprise management, finance, personnel and the like, can cover more possibly used crowds, and can make important basis for later user crowd use types.

The method comprises the steps of dividing billions of data into concepts, attributes and entities through graph calculation to form structured data, finally displaying the structured data in the form of knowledge cards, meeting the requirements of questions and answers in the user material field, quickly searching answers in the material field from massive data, enabling the content of the knowledge cards to cover concepts of different types of data, such as products, raw materials, industries, personnel, institutions and the like, and corresponding attribute relations, establishing a structured data structure to form knowledge cards of different types, such as accurate questions and answers, comparison types, reasoning judgment types and the like, and enabling the most accurate card results to be arranged at the front position when users search results.

Meanwhile, according to the billions of data in the database, the content extraction in the material field can be carried out, according to the corresponding templates, field data in data corresponding to products, industries, raw materials and the like are automatically generated and selected, the data are automatically matched to form the data report products corresponding to tens of millions of data, and the PDF report products of the latest dynamics, supplier recommendation, expert recommendation and technical development process at home and abroad in the material field are generated, so that the method is continuously expanded and enriched and is used for teachers, students, experts in the material field and enterprises in related research institutes to read and download online.

2) The recommendation is applied to the overall search results, affecting the location and ranking of the search results. According to industry experience, user data and the like, potential rules in the data are actively found, a recall and sorting algorithm is optimized according to cold start information, a recommendation effect is iterated rapidly, and thousands of persons and thousands of faces are individually recommended. And supporting the comprehensive consideration rules based on the authoritative issuing text such as new material application demonstration catalogue', material dictionary, user portrait label, search keyword, multi-level problem and other factors as recommendation. The index of the recommendation algorithm may be determined according to related indexes (such as document access amount, download amount, influence factor), the popularity of the keywords (whether the keywords are core keywords), the keyword frequency, and the like.

3) The cognitive search is realized by a search technology based on a graph calculation engine, and the mode of returning the current search request based on the user to the related Web page links of the user brings great convenience to the internet information retrieval. However, the mode still has the defect that the form of the result returned by the search is single, accurate information can not be directly provided based on the search request of the user, and the user still needs to search the required information in the webpage according to the link provided by the user.

The user wishes to search for information that can directly give the relevant Entity (Entity) and Entity information related to that Entity according to his search request. If the user searches for "electronic science and technology university", the user is more likely to want to acquire the information of the school distribution, professional distribution, address of the school, contact way, historical famous alumni, school, etc. of the school; rather than just the relevant web page links.

Searching requires further entity information mining of the prior knowledge and representing the knowledge of the part in a good structure and indexing the mined knowledge efficiently. The user can quickly and conveniently search the most needed information based on the knowledge of the part, and on the basis, the semantic relation between related entities can be provided, and finally a knowledge network between results is formed.

The cognitive search module mainly comprises knowledge card search, chart card search, exploration analysis, index customization, search analysis, search prompt, chart function, source data analysis, reasoning search, accurate search, data search, webpage data search, picture data search and other functions.

The technical layer is that all the entities are obtained through entity knowledge fusion and entity alignment, a knowledge graph is formed by utilizing semantic relations among the entities, the knowledge graph completes information request of a retriever, and related information is recommended and a final ordering result is given through analysis of the retrieval result. According to the scene of the service, defining a schema, extracting a relation, defining a knowledge type topic, extracting service definition, and supporting semantic recognition and understanding based on the result of map construction.

The result forms of the cognitive search are divided into content results, knowledge card results and chart analysis results, wherein the content results are all content types related to the material field at present, and include information web pages, patents, journals, reports, pictures, videos, reports and material analysis report contents generated manually and automatically in the deep search.

The knowledge card results are constructed and designed based on the atlas, and according to the association relation between the edges and the concepts generated by the material attribute relation, the formed reasoning, figures, various institutions of products and other cards can be used for carrying out the question and answer reasoning results in the form of the card through the question of the user on the material attribute, for example, the question of whether polyimide is an organic matter or not is given out, the accurate results of yes and no are given out through the knowledge card reasoning capability during searching, when the user wants to inquire about a university or a college in the material field, the relationship between the concepts and the attributes can be also used for giving out the card results of figures or a university institution and the like, corresponding answers are given out at first, the time of the user for inquiring and screening the questions through mass content is reduced, the user requirements are met, and the user difficulty of the material figures is solved efficiently and intelligently.

The manual and automatic generation of the material analysis report product is realized by combining the corresponding field of the database with the corresponding report management background of the product design, customizing the template and automatically taking the number, and generating the corresponding material analysis report. The report is from the material field database, according to field data and template text required by material field personnel to comb, the corresponding analysis ten-thousand-level number report is automatically generated by analyzing and matching the current material name and industry of the database, the content is not only limited to the text, the chart, the picture and other forms, more accurate and credible research conclusion support is provided for the material field personnel, the complicated data analysis is more highly efficient and convenient, the complex process of manually writing and automatically analyzing the data to generate the report in the past is changed, and the difficulties and bottlenecks in writing report analysis by the current material field personnel are effectively solved.

4) The map management module mainly comprises sub-functions of visual map management, map display, version management, map design and the like, manages the current design through designing and classifying the maps, can check the number and specific information of each entity and whether the entities are associated with each other or not, and can modify the maps, such as conceptual relationships, attribute relationships, entity relationships and the like, so that the maps are more accurate and are more fit for business requirements.

5) Recommendation ordering algorithm

In the functional aspect, potential rules in the data are actively found according to industry experience, user data and the like, recall and sequencing algorithms are optimized according to cold start information, a recommendation effect is iterated rapidly, thousands of individuals and thousands of individuals are recommended, and content requirements of different crowds are met.

And supporting the comprehensive consideration rules based on the authoritative documents such as material dictionary, user portrait labels, search keywords, multi-level questions and other factors as recommendations. The index of the recommendation algorithm may be determined according to related indexes (such as document access amount, download amount, influence factor), the popularity of the keywords (whether the keywords are core keywords), the keyword frequency, and the like.

In the technical aspect, recall based on similar behaviors and recall based on similar contents are supported; special weighting for recall edges is supported, and weighting configuration can be carried out without complex configuration; the construction of a first order model on a recall set is required to be supported; and the construction of a secondary sorting model is supported after the recall of the collection or the primary sorting result.

And when the recommendation is established, the recommendation is comprehensively carried out according to recommendation requirements, requirement analysis/disassembly, sample set collection, feature selection, recall model, sequencing model, special rules, secondary sequencing model and other recommendation consideration. Meanwhile, the method comprises a shallow migration model, a deep migration model, a collaborative filtering model, a Bayesian recall model and the like under the framework of the graph calculation engine. Including arbitrary path walk computation under the graph computation engine framework, collaborative walk, point mutual information, pearson, etc. Feature screening of graph structures; and designing a map by supporting a visualized map structure, and selecting a characteristic candidate set on the visualized map structure.

6) Graph calculation search analysis module

Based on the knowledge graph technology, the real-time construction of indexes and the exploration of data in multiple dimensions and layers are realized through a real-time graph calculation engine. The core function points of the graph calculation engine have high performance and real time, such as calculating attribute distribution of more than millions of potential entities, only needing second-level return results, and meanwhile, the graph calculation engine has the characteristic of flexibility, highly abstracts various relations, and can well express various abstract relations such as user behaviors, entities, entity attributes and the like. Precision: the method can directly support models such as real-time clustering, correlation analysis, attribution analysis, bayesian network and the like, and provides accurate prediction service oriented to business logic.

Traditional relational data expose modeling defects, lack of real-time and other problems in ROLAP (multi-entity relational online analysis), so that a graph calculation engine with more powerful expression capability is adopted to support complex relational calculation. It has the following advantages:

high performance and real time: the attribute distribution of over a million potential entities is calculated, and only the second level is needed to return results, which is 60-1200 times that of the traditional database query.

Flexibility: any relationship type can be dynamically inserted, and the method is very suitable for storing dynamically-changing information (such as human behaviors).

Abstraction: various relations are highly abstract, and various abstract relations such as user behaviors, entities, entity attributes and the like can be well expressed.

The adjacency matrix of the graph is realized as a typical graph storage structure, n vertexes are stored in a one-dimensional array, and the relation between any two points is represented by a matrix of n. The main diagonal is all dead space and the sum of the number of row and column edges of the vertex represents its degree of egress and ingress, respectively. It is apparent that for sparse matrices with a small number of edges relative to vertices, memory space is wasted significantly. The adjacent table stores the vertexes by using a linear table, and the adjacent vertexes form a linear table additionally, so that a linked list storage is often used because the number of adjacent points is not determined.

There are two types of graph splitting, edge splitting and point splitting. The edge segments are stored once for each vertex, but some edges are broken down onto two machines. The benefit of this is that storage space is saved; when the graph is calculated based on edges, for edges with two vertexes separated on different machines, data is transmitted in a cross-machine communication mode, and the communication flow of an intranet is large. The point cut is stored only once per edge and will only occur on one machine. The multiple adjacent points are copied to multiple machines, so that the storage overhead is increased, and meanwhile, the problem of data synchronization is caused. The method has the advantage that the traffic of the intranet can be greatly reduced.

And dividing the graph data, and distributing the data to different nodes in the cluster. The data quantity distributed to each node is uniform as much as possible, and the phenomenon of unbalanced load caused by that a large amount of data is inclined at some nodes is avoided. The Hash partitioning takes the remainder, i.e., hash (Key)% M, for the Hash result for each point, assuming that the machine numbers from 0 to N-1, modulo M according to a custom Hash () algorithm, the Hash () value for each request, obtaining the remainder i, and then distributes the point to the machine numbered i. The dots can be evenly distributed over M machines.

The knowledge graph calculation engine can directly support models such as real-time clustering, correlation analysis, attribution analysis, bayesian network and the like, and provides accurate prediction service oriented to business logic. And the method is based on the capability of a graph calculation engine, so that the accuracy of the cognitive search result and the knowledge recommendation with high correlation are realized, and the analysis and the reasoning of complex knowledge are performed.

More specifically:

the material vertical search engine realizes various product functions through 9 large implementation examples, including multiple rounds of machine question and answer, cognitive search, data application center, intelligent report, map management, knowledge management, map calculation search analysis module, data docking and user portraits, and the following are specific details:

Multiple rounds of machine questions and answers: based on the knowledge graph technology, the semantics of the public natural language questions can be accurately understood, and knowledge answers can be accurately pushed. The PC end and the mobile end are supported to use the functions.

Cognitive search: the analysis index self-defined configuration is supported, the search intention of the user can be intelligently understood, the multidimensional data analysis result conforming to the question is positioned, and the graphic display of tables, bar graphs, pie charts and the like is supported.

And the data application center: providing data insight report auto-generation, application market display, and subscription functionality.

Intelligent reporting: report template management, report generation, and report download functions are provided.

And (3) map management: the knowledge graph management function comprises visual graph management, graph data access, graph display and the like.

Knowledge management: providing knowledge classification management, knowledge template management, knowledge flow management and other functions.

The graph calculation search analysis module: knowledge data based on the knowledge graph and a graph calculation engine can realize the calculation engine of the knowledge graph and support the capabilities of cognitive search, knowledge recommendation, flow reasoning and association analysis.

And (3) data docking: and a data connection tool is provided to realize the butt joint of webpage data, picture data, data, product data and the like.

User portrayal: the user tag system function is provided, data acquisition is carried out on basic information and other information of a user in the user registration stage, and data acquisition is carried out on searching, clicking, staying, collecting and other behaviors of the user in the user use stage, so that user portraits are formed.

As shown in fig. 2, the overall structural design includes:

the material search engine product integrally comprises four layers, namely an application layer, a capability layer, a calculation layer and a data layer. The user portrait is formed by collected user basic information, user behavior and other data, product data, industry trend prediction, client demand prediction, literature summary report, industry research report and other data required by intelligent recommendation are provided by multiple rounds of machine question-answering and user interaction, and transaction flows such as shopping cart adding, payment and the like can be completed on line in one stop aiming at payment data.

Meanwhile, the method supports exploratory analysis of data, and through analysis index custom configuration, the searching intention of a user can be intelligently understood, a multi-dimensional data analysis result conforming to a question is positioned, and graphic display such as a table, a histogram, a pie chart and the like is supported.

As shown in fig. 1, the functional architecture includes:

the material vertical search engine is a special search product in the field of material science, and the functions of the material vertical search engine are designed around the material science, including analyzing user behaviors, material knowledge questions and answers and solving massive data analysis results, so that the academic knowledge problem of material science personnel is solved.

The product functional design comprises user portraits, multiple rounds of machine questions and answers, cognitive searches, data application centers, intelligent reports, map management, intelligent reports, knowledge management, graph calculation engines and underlying data interfacing.

As shown in fig. 3, the technical architecture includes:

1. the presentation layers of cognitive searching, map management and the like are based on front-end frameworks (VUE) which are popular at present, and front-end and back-end separation is achieved.

2. The spectrum management, cognitive search and other services are realized based on a springboot architecture, and the rest interface mode provides services.

3. The recommended algorithm service is based on the python nlp related technology framework, and the rest interface mode provides service.

As shown in fig. 4, the deployment architecture includes:

1. the user accesses the platform front-end service through load balancing, which can be a nginx load or an F5 hard load.

2. The front-end service of the modules such as multi-round machine question-answering, cognitive searching and the like is deployed on the front-end server, and multi-node deployment is performed.

3. The back-end service of the modules such as multi-round machine question-answering, cognitive searching and the like is deployed on a back-end server, and multi-node deployment is performed.

4. The ETL service is independently deployed and is responsible for data butt joint with an external system. The data received is first stored in kafka and then distributed by kafka for storage in the graphics engine and ES index.

5. mysql is deployed master-two-slave.

6. redis caches a master-slave deployment.

7. The graph computation engine is deployed with 2 nodes.

8. ES indexes 3 node deployments.

Example two

The embodiment discloses an intelligent search engine method in the vertical field of materials.

It will be appreciated by those skilled in the art that the modules or steps of the invention described above may be implemented by general-purpose computer means, alternatively they may be implemented by program code executable by computing means, whereby they may be stored in storage means for execution by computing means, or they may be made into individual integrated circuit modules separately, or a plurality of modules or steps in them may be made into a single integrated circuit module. The present invention is not limited to any specific combination of hardware and software.

While the foregoing description of the embodiments of the present invention has been presented in conjunction with the drawings, it should be understood that it is not intended to limit the scope of the invention, but rather, it is intended to cover all modifications or variations within the scope of the invention as defined by the claims of the present invention.

Claims

1. A material vertical domain intelligent search engine system, comprising:

2. The materials vertical domain intelligent search engine system of claim 1, further comprising a cognitive search module, wherein:

3. The materials vertical domain intelligent search engine system according to claim 1, wherein the user's interaction comprises searching, clicking, staying, collecting, and the usage scenario comprises information, pictures, patents, journals, and reported search pages.

4. The intelligent search engine system in the vertical material domain according to claim 3, wherein the graph calculation search analysis module is further configured to set the questions with high matching degree to the user in a multi-level form and set weights of various levels according to a plurality of influencing factors, and when pushing to the multi-round question-answering module, the questions are pushed with priority according to the weights of the multi-level questions, wherein the plurality of influencing factors include whether there is data, commercial value of the data, and hot spot of the material.

5. The materials vertical domain intelligent search engine system of claim 2, wherein the graph computation search analysis module is further configured to generate a materials analysis report by parsing a user search intent, matching the user search intent with a current materials name and industry in a materials domain database, obtaining template text and related field structured data from the materials domain database, and expanding and enriching the report.

6. The intelligent search engine system of the vertical field of materials according to claim 1, wherein the graph calculation search analysis module is further used for responding to the question of the user for the attribute, generating the association relation between the edge and the concept according to the attribute relation based on the graph calculation method, and forming knowledge card results of each organization of reasoning, people and products.

7. The materials vertical domain intelligent search engine system according to claim 1, wherein the graph analysis results support a plurality of graphical presentations including tables, bar graphs, pie charts.

8. The intelligent search engine system in the vertical material field according to claim 2, wherein the graph calculation search analysis module is used for understanding the user search intention and the user question and answer intention based on a graph calculation method and performing real-time clustering, correlation analysis and attribution analysis to obtain a data analysis result and a predicted answer.

9. The materials vertical domain intelligent search engine system of claim 2, further comprising a user portrayal module, a profile management module, an intelligent reporting module, a knowledge management module, a data docking module, and a data application module, wherein:

10. An intelligent search engine method in the vertical field of materials is characterized in that: the method comprises the following steps: