Task Based Design Knowledge Collaborative Filtering Recommendation and Its Application

Full text


2017 International Conference on Mathematics, Modelling and Simulation Technologies and Applications (MMSTA 2017) ISBN: 978-1-60595-530-8

Task Based Design Knowledge Collaborative Filtering

Recommendation and Its Application

Yong-jian ZHANG


, Yang-fan XU and Hui-yue CI

School of Naval Architecture and Ocean Engineering Harbin Institute of Technology (Weihai), Weihai, Shandong, 264209, China

*Corresponding author

Keywords: Knowledge recommendation, Collaborative filtering, Design process.

Abstract. The overwhelming amount of knowledge discovered form product data necessitates mechanisms for efficient information filtering during complex product design. In order to overcome the incompatibility of the traditional collaborative filtering algorithm in product design knowledge recommendation, A process-centric design knowledge management model is given, in which knowledge is reasonably classified and effectively organized. A task based knowledge recommendation algorithm considering task similarity and time factor is given. Application of the model and algorithm on construction machinery engine design support system is introduced and discussed.


Nowadays time for design and production need to be decreased to increase the enterprise’s rapid response capabilities and competitiveness. Mining implicit knowledge from history design data become more and more important [1]. Meanwhile, with the increasing of knowledge amount, how to use appropriate knowledge to assist given designer to accomplish specific task becomes a new challenge in knowledge management [2]. The overwhelming amount of data necessitates mechanisms for efficient information filtering. Collaborative filtering is one of the techniques used for dealing with this problem. The recommendation technology has become a hot area in both academia and industries [3]. Collaborative filtering algorithm (including user-based method, item-based method, fusing user-based and item-based method [4] and their extensions [5, 6]) with its easy to implement and good scalability became the most popular recommendation algorithm. However, traditional recommendation algorithm does not consider contextual information and are difficult to apply to process-sensitive complex product design issues [7, 8]. Based on a process-centric design knowledge management model, a task based deign knowledge recommendation algorithm and its application on construction machinery engine deign purport system is discussed in this paper. This paper is organized as follows: In the next section, process-centric design collaboration model is given in detail, and based on this model, the classification and organization of design knowledge are discussed. Next, design knowledge recommendation algorithm based on traditional user-based collaborative filtering algorithm by taking into account of design task information is introduced. In application section, a design support system for construction machinery engine is discussed. In the last section, the paper is concluded.

Process-Centric Design Knowledge Management Model


Product design knowledge is divided into two major categories: dynamic knowledge and static knowledge. Dynamic knowledge describes dynamic product design process DP which consists of several design tasks DT and their relationship TR, denoted as DP=<DT, TR>. In order to improve the reusability of design process, design process model DPM is built by abstracting historical product design instances which are recorded and stored in databases of Product Data Management systems. Product design process model also consists of a number of design tasks. To distinguish between tasks in design process model and those in design process, the former is referred to as standard task ST in this paper. Standard task ST can be described in detail as ST = <IN, OUT, DK, DR). In ST, IN is design task input which is the basis of current design work. OUT is design task output, which is the actual deliverable of design activity and can be used as input for subsequent design tasks. DK is design knowledge that helps designers get output from input. DR=<A, C> is design requirement that express constraints and goals of design task, where A is product attribute and C is attribute value constraint. Static knowledge refers to the design knowledge that does not contain procedural information, such as design manuals, design cases, etc. According to the difference of knowledge expression, static knowledge can be divided into two types of structured knowledge and unstructured knowledge. Structured knowledge uses formal methods for expressing on the basis of knowledge modeling, for example, product design cases are usually described by attribute-value pairs. Unstructured knowledge, such as texts, images, etc., is more suitable for reuse by dedicated software or manual methods.

According to the classification and definition of knowledge, the design process model plays a role of knowledge organization. The standard task in the design process model is the carrier of static knowledge. Instantiating the design process model can construct design process for new product and transfer various kinds of knowledge to different design scenarios and designers. For a given design task, input, output, and design requirements need to be set in advance. Design knowledge often needs to be queried and acquired by designer or application based on the design task’s intent. Detailed classification of static knowledge and its association with standard tasks can reduce the blindness of knowledge acquisition in the design process. However, due to the explosive growth of design knowledge, application of design knowledge recommendation technology based on the knowledge management model will further improve the quality of knowledge reuse.

Standard Design Task i


Standard Design Task k




Standard Design Task j


… …


Documents Duk

Design Cases Dck

Design Requirements DR={A, C}


Figure 1. Product design process model.

Task Based Knowledge Recommendation Algorithm

Given design task Tk{ ,T1Tn} where n is the number of tasks in task set with the same standard task type, designer Ui{ ,U1Ul} is executor of Tk and l is the number of users in user set. The knowledge recommendation problem is how to recommend useful knowledge set { } { ,KK1Km} for user Ui to finish task Tk where m is the number of knowledge elements in knowledge set.


design activity. For the same user, different design tasks required different design knowledge. There are also differences between design knowledge required for design tasks in the past and knowledge required for now. To solve design knowledge recommendation problem, this paper made improvements on traditional user-based collaborative filtering algorithm by taking into account of design task information.

First, by mining design intent similarity using design task information, the range of knowledge available for reference will be narrowed down, knowledge dimension in the user-knowledge matrix will be effectively reduced, then efficiency and accuracy of design knowledge recommending algorithms can be improved. The design task’s input and output such as manuals drawings and 3D models are unstructured in nature and difficult to calculate. So, given design task T Tp, q{ ,T1Tn}, design intent similarity sim T T( , )p q was measured by calculating the set similarity of task design requirements. Without loss of generality, constraint value of the uth requirement attribute p

u A for design task Tp are represented by trapezoidal fuzzy number p ( p,1, p,2, p,3, p,4)

u u u u u

C  c c c c , where

,2 ,3

[ p , p ]

u u

c c represents the expected value interval of attribute p u

A and [ p,1, p,4]

u u

c c represents the limited value interval of attribute p


A . The calculation of sim T T( , )p q is shown in Eq. 1. In this equation,

p, q

u u

dist C C  is the distance of constraint value p


C and q u

C which is shown in Eq. 2, ( ( )) u

range C  is

maximum value range of uth requirement attribute for all design task which is shown in Eq. 3.

( ) ( )

( , )= ( ( ) p, q ) / ( )

p q u u u u


sim T T

range C  dist C C  range C 


4 2 1/2

, ,


1 ,


p q p q

u u u u

dist C C c c

   

   (2) ( ) ,4 ,1 1 1

( )= ( v ) ( v )

u v n u v n u

range C max c min c

  

 

(3) For executor Ui of design task Tk , knowledge from tasks that are similar to Tk is used for recommendation. Given a threshold ε, knowledge set { ,K1Km} consists of Documents and Design Cases of all the tasks in task set { } { |TT sim T Tu ( , )k u }, user set { ,U1 Ul} consists of executors

of all the tasks in task set { }T . Then user-knowledge rating matrix R l m( , )

 

rik l m can be established from user set { ,U1Ul} and knowledge set { ,K1Km}, in which rik is rating value from user Ui for knowledge element Kk, rik can be set by the user after design task is performed.

For product design, with the continuous improvement of design technology and accumulation of designers' experience, the reference value of knowledge will gradually decline. Considering time factor of design task, user similarity expression based on Pearson Correlation Similarity is as follows.

2 2 ( )( ) ( , )= ( ) ( ) iv iv iv

ij tj i vj tj v

j K

i v

ij tj i vj tj v

j K j K

r w r r w r

sim U U

r w r r w r

      

(4) In the above equation, wtj is a function of difference between knowledge Kj most recently used time of and current time, and a larger value of wtj indicates a higher credibility of knowledge Kj


set using the following simplified model: wtj 1 when t ttrust , wtj 0 when t tuntrust , wtj

decreases linearly from 1 to 0 when ttrust  t tuntrust.

Based on the user similarity measure, a set of K-nearest-neighbor users KNN U( )i of user Ui can

be obtained. Using user similarity as weight, rating value rij is the weighted average of rating scores of knowledge Kj from all users in K-nearest-neighbor users KNN U( )i .

( )

( )

( , )( )


( , )



i v vj tj v


ij i

i v


sim U U r w r

r r

sim U U

 

(5) Finally, top-N knowledge elements were served as a reference for user Ui to finish task Tk. Algorithm flow of task based knowledge recommendation as shown in Fig. 2.

Design Process Logs New Task User


Similar Tasks

Users Knowledge

1 U






K KjKm

Rating and Sorting Time Factor


Figure 2. Task based knowledge recommendation process.

Application and Discussion

Construction machinery engine is a typical complex product, its design process covers the entire product lifecycle, from concept selection, customer needs analysis to production validation and release. A design support system needs to be established to manage the entire process of product design and data and knowledge generated during the process. New product design projects (including tasks in the project, the transfer of information between tasks) need to be automatically created by the system according to product design requirements. During the design process, designers only need to be concerned with their own design tasks. When upstream design tasks give all the inputs needed for the current design task, the design task starts automatically and the designer needs using the knowledge that system managed (such as manuals, patents, historical cases, competing products, etc.) to deliver the required output of the task. A web-based engine design support system is developed as shown in Fig. 3. In this figure, the tree structure corresponds to the engine design process model, the leaf nodes of the tree are standard tasks. Engine design support system has been integrated with PDM system, competing products management system, patent knowledge management system. Documents, 3D models, etc. in PDM system can be effectively reused in the design project of new products.


documents. Finally, there is still a problem of poor accuracy in the evaluation of knowledge by log mining, used knowledge needs to be manually evaluated by designer after finishing the task, therefore, new and more effective knowledge rating methods need further study.

Figure 3. Design support system for construction machinery engine.


We are grateful to the reviewers and the editors for their constructive suggestions. This work was supported by the National Natural Science Foundation of China (Grant No. 51405104).


[1] Z. Wu, J. Liao, W. Song, H. Mao, Z. Huang, X. Li, H. Mao, Product lifecycle–oriented knowledge services: Status review, framework, and technology trends, Concurr. Eng. 25 (2017) 81–92.

[2] D.S. Li, Q. Lv, L. Shang, N. Gu, Item-based top-N recommendation resilient to aggregated information revelation, Knowledge-Based Syst. 67 (2014) 290–304.

[3] F.O. Isinkaye, Y.O. Folajimi, B.A. Ojokoh, Recommendation systems: Principles, methods and evaluation, Egypt. Informatics J. 16 (2015) 261–273.

[4] B. Wang, J. Huang, L. Ou, R. Wang, A collaborative filtering algorithm fusing user-based, item-based and social networks, in: 2015 IEEE Int. Conf. Big Data (Big Data), 2015: pp. 2337–2343. [5] Y. Ar, E. Bostanci, A genetic algorithm solution to the collaborative filtering problem, Expert Syst. Appl. 61 (2016) 122–128.

[6] M. Maleszka, B. Mianowska, N.T. Nguyen, A method for collaborative recommendation using knowledge integration tools and hierarchical structure of user profiles, Knowledge-Based Syst. 47 (2013) 1–13.

[7] C.H. Lai, D.R. Liu, Integrating knowledge flow mining and collaborative filtering to support document recommendation, J. Syst. Softw. 82 (2009) 2023–2037.


Figure 1. Product design process model.

Figure 1.

Product design process model. p.2
Figure 2. Task based knowledge recommendation process.

Figure 2.

Task based knowledge recommendation process. p.4
Figure 3. Design support system for construction machinery engine.

Figure 3.

Design support system for construction machinery engine. p.5