Research On Patent Knowledge Mining Method And Its Prototype System To Support Product Conceptual Design

Conceptual design is the core of product innovation,which is a knowledge-driven problem-solving process.Its essence is the process of knowledge transfer and reorganization.As a product of innovative design,patent document has become a new knowledge resource to stimulate designers to obtain innovative inspiration,which is most closely related to product innovation activities.Therefore,how to mine patent knowledge from massive patent documents to assist conceptual design has become the key to product innovation design.The classification and retrieval of patent documents is not only a necessary tool for patent knowledge mining,but also an important research content to support product conceptual design.This dissertation takes Chinese patent documents as the research object,and the purpose is to acquire the knowledge required for product conceptual design.From the perspective of patent document classification and patent knowledge retrieval,research on some of the key issues and methods and tools that support conceptual design,and a new method of patent knowledge mining supporting product conceptual design is proposed.The main contents of this dissertation are as follows:(1)According to the related research of the existing conceptual design solution process model,the design element information that constitutes the conceptual design solution process model is analyzed,and then the Requirement-Function-PrincipleStructure(RFPS)process model of patent knowledge aided product conceptual design is constructed.The functional basis is established to standardize the expression of the functions in the RFPS model and the roles and relationships of each layer in this model are discussed in detail.(2)Taking several existing term weighting methods as examples,firstly,this paper analyzes and expounds how to reasonably calculate the term weights for test set patents in patent classification,and on this basis,five unsupervised term weighting methods are proposed.Then,aiming at the shortcomings of the current supervised term weighting method in patent classification,a multi-level patent term weight calculation model integrating term-text-category is established.And on this basis,a supervised term weighting method is developed based on cumulative residual entropy.Finally,aiming at the problem of unsupervised and supervised term weighting methods that do not consider the semantic relationship between terms,the weighted term vector is constructed,and then the evidence theory,circular convolution and convolutional neural network are introduced to complete the synthesis of weighted term vector.Furthermore,in order to solve the sparse problem of traditional vector space model,a semantic-based vector space model is proposed to realize the term weight calculation of patents.The feasibility and effectiveness of the proposed method are verified by experiments.(3)According to the characteristics of design knowledge contained in patents,the verb+noun pair,verb and noun are used to represent function knowledge,principle knowledge and structure knowledge respectively,and hence the representation of patent knowledge is completed.Taking Stanford Corenlp as part of speech tagging tool to elaborate the rules of patent knowledge extraction.Aiming at the knowledge of drawings in patents,the algorithm of extracting patent drawings from patent is given.A patent knowledge organization and storage strategy with patent title and abstract as index is established,which provides support for further patent knowledge retrieval.(4)As one of the necessary tools to use patent knowledge,patent knowledge retrieval determines the efficiency and effect of patent knowledge application(i.e.,transfer and recombination).In view of the phenomenon of using synonyms in keyword retrieval methods and the problem that keywords can not fully express the retrieval intention,the intention unit is established according to the International Patent Classification(IPC)to identify and understand the retrieval intention.A patent knowledge retrieval method based on IPC is proposed,and then the calculation process of this method is described in detail.Considering that the number of functional verbs contained in each intent unit and each patent cannot be exactly the same as each other,a complete set of functional verbs used to describe actions is extracted from the IPC and the patent respectively,and then the absolute index of the function used for patent knowledge retrieval is defined.It solves the problem of using only cosine function to calculate similarity in patent knowledge retrieval,and improves the accuracy of patent knowledge retrieval.(5)Based on the aforementioned theoretical research,a patent knowledge mining prototype system(PKMPS)supporting product conceptual design is developed.The PKMPS is applied to the innovative design of the irradiated sample hole plug of the pressurized water reactor.The window layout of each functional module of the PKMPS is displayed,and the operation rules and usage methods of each functional module are introduced in detail.The feasibility and practicability of the PKMPS in aided product conceptual design is preliminarily verified.