Research Progress Kieu Que Anh School of Knowledge

  • Slides: 19
Download presentation
Research Progress Kieu Que Anh School of Knowledge, JAIST

Research Progress Kieu Que Anh School of Knowledge, JAIST

Outline • We present the current design creativity model of (Li et. al, 2007)

Outline • We present the current design creativity model of (Li et. al, 2007) and how it link to our work • We present our current works on the research – The status of opinion summarization – How to collect data (what kind of reviewer) • Strategy for selecting concepts using the Word. Net similarity

Creative Design model [1] According to the psychologist Sternberg , personal creativity is related

Creative Design model [1] According to the psychologist Sternberg , personal creativity is related to a person’s intelligence, knowledge, thinking styles, personality, motivation, and environmental context. Therefore, it can be expressed as, For product innovation K : knowledge I : intelligence TS: means thinking style P: personality M: motivation E : environment context DM is design method ST is computer supporting tool U is uncertainty factor KB is knowledge based IR is information resource CAT is computer aid thinking

Design creativity model [1] The authors emphasis the contribution of information technology Concept operators

Design creativity model [1] The authors emphasis the contribution of information technology Concept operators Knowledge base and information Resource can be used with tools Can we use concept operators here?

My work and the link to this model • Using design creativity for product

My work and the link to this model • Using design creativity for product innovation • Using opinion extraction to support designer • The link to design creativity model[1]? – Opinion extraction likes ST part in the creative product innovation model – Concept operators is DM part in the model

Our Model Obtain useful opinion text Obtain a list of products and their concepts

Our Model Obtain useful opinion text Obtain a list of products and their concepts Step 3. Creating new concept in design creativity Word. Net Sentiment Word. Net

Concept operators Collecting base concepts Select appropriate set of concepts Applying concept operators •

Concept operators Collecting base concepts Select appropriate set of concepts Applying concept operators • Collecting base concepts from outputs of opinion extraction – Product’s names – Positive features of the products – Negative features of the products • Selecting an appropriate set of concepts using Word Net – Two dissimilar concepts are confidently selected (Taura et. al, 2005) – A concept with a greater number of association concept are confidently selected as base concepts (Nagai and Taura, 2006) • Applying concept synthesizing, concept blending, analogy for two base concepts

Collecting base concepts • Collecting concepts based on the main webs – Amazon. com

Collecting base concepts • Collecting concepts based on the main webs – Amazon. com and eopion. com, ebay. com – Indexing all product names and their attributes on a collection (such as dictionary) – Currently, we collect some types of products including: • Ipod, mobile phone, Iphone, camera ( Digital Cannon).

Collecting negative and positive attributes Extract pros and cons features which are available on

Collecting negative and positive attributes Extract pros and cons features which are available on eopinion. Using opinion extraction to extract more useful information about features of product

Our strategy for collecting review opinion – Those opinion documents from “top reviewers” •

Our strategy for collecting review opinion – Those opinion documents from “top reviewers” • Top Reviewers help customers find the best products on Epinions (http: //www 0. epinions. com/) by writing high quality reviews in their category of expertise. • Top Reviewers' reviews have received the highest ratings from Advisors in their category and the community at large. – The opinion documents from normal reviewers – We extract “pros” and “cons” information in each opinion documents for positive and negative features

Opinion Extraction Opinion Text Topic Identification Opinion Summarization opinion summary • The opinion extraction

Opinion Extraction Opinion Text Topic Identification Opinion Summarization opinion summary • The opinion extraction method of (Zhan et. al, 2008) is implemented to extract useful information for a given opinion text. – We obtain a topic for each document using topic identification model – We use the Maximal Marginal Relevance method for opinion summarization • Its idea: A sentence is included in a summary if it satisfies – Similar to the topic – Different from those sentences already in the summary. • Current status: we have finish the tool and now evaluate its outputs – Our system can generate • A set of user’s opinion document is summarized • A list of positive features and negative features for each product

Current status of opinion extraction • • • Topic Identification: It detects phrases appearing

Current status of opinion extraction • • • Topic Identification: It detects phrases appearing in many review documents We obtain all phrases which are frequently in the review document collection Temporary results (<score> confident </score>; <what> topic </what>; <where> locations (i. e document index) </where>) How to compute confident score? <pattern> <score>22. 2026</score><what>cell phone </what> <where>8 20 35</where> </pattern> <score>22. 2026</score><what>flip phones </what> <where>16 24 31</where> </pattern> <score>26. 5754</score><what>sound quality </what> <where>8 13 18 20 23 27</where> </pattern> f (topic) is frequency N is number of document n is number of document f appear l is the length of f

Opinion summarization • Using MMR (Maximal Margin relevant) method for summarizing a set of

Opinion summarization • Using MMR (Maximal Margin relevant) method for summarizing a set of documents: • For example: (the topic “sound quality” can be summarized using a set of documents defined by topic identification step: reception[+], sound quality[+] the reception and sound quality are top-notch. i can routinely talk on the phone in my house in the suburbs , which is an unaccustomed luxury , even with t-mobile service. size[+], weight[+] i like the size and weight of this little critter This example is obtained by performing on Nokia product review

System’s output • We obtained a data of reviewer from eopinon. com and amazon.

System’s output • We obtained a data of reviewer from eopinon. com and amazon. com – The product for selecting output include: Ipod and Nokia phone • System’s output – “topic” <sound quality> are mentioned frequently in reviewer’s comments. – We then exact reviewer’s comments related to this topic. This information will be provided for designers. <pattern> <score>26. 5754</score><what>sound quality </what> <where>8 13 18 20 23 27</where> </pattern

System’s output for top reviewer’s comments • We found that in top reviewer’s comment,

System’s output for top reviewer’s comments • We found that in top reviewer’s comment, they usually indicate the advantage and disadvantage of products. • The system can extract “advantage” and “disadvantage” information using some key-phrase like (cons, pros, “drawback”, etc) • EX: <top reviewer: Chris. Joker's Full Review: Apple i. Pod touch (16 GB, MA 627 LL/A) > – Pros: Great wireless applications, easy to use, video screen is great – Cons: Quick battery depletion, less memory than other i. Pods

Select two concepts • We use Word. Net similarity for selecting two concepts –

Select two concepts • We use Word. Net similarity for selecting two concepts – The tool can be obtained at: http: //wn-similarity. sourceforge. net/ – We also can use this tool online – It is used to measure the similarity between two concepts: we select the path-based method

Word. Net Similarity Root 2 measuring “closeness” of concepts in terms of their definitions

Word. Net Similarity Root 2 measuring “closeness” of concepts in terms of their definitions using LCH method D=5 Medium of exchange 1 Money 1 1 Credit 1 Cash Lch Related (Money-Credit) = -log (2/10) = 0. 70 1 Credit card Coin 17

Example • Cell phone – – – • A Phone is controlled by hand

Example • Cell phone – – – • A Phone is controlled by hand written Pen – – • Sharp Color Screen Size control The similarity of phone and pen using Word. Net is 0. 1427. Sharp Color Size write The similarity of phone and youngster 0. 0909 youngster -travel a lot -young A Phone with flexible form

Question and Answer • Thank you very much for your attending!

Question and Answer • Thank you very much for your attending!