+ Argumentation-based Discussion for User Forum: A Research - PowerPoint PPT Presentation

+ Argumentation-based Discussion for User Forum: A Research Preview Itzel Morales Ramírez and Anna Perini REFSQ 2014 Essen, Germany, April 2014

+ Outline 1 n Context + Motivation n Problem + Research questions n Ideas + Research plan n Conclusion Argumentation-based Discussion for User Forum: A Research Preview Itzel Morales Ramírez Thursday, April 17, 2014

+ Context + Motivation 2 n User forums provide a virtual space in which participants exchange views on issues about a software application. n Open source software (OSS) projects exploit this idea. n Users and volunteer developers of OSS communities engage in discussions derived from initial requests. n Data from: Apache OpenOffice (AOOo) bugzilla + Argumentation-based Discussion for User Forum: A Research Preview Itzel Morales Ramírez Thursday, April 17, 2014

+ Problem 3 n Tedious discussions to be completely read n Sequential comments that hide an explicit recognition of participants’ attitude (i.e. “in favor” or “against”) wrt. the initial request. n Unclear identification of worth comments to be further analyzed 3 Argumentation-based Discussion for User Forum: A Research Preview Itzel Morales Ramírez Thursday, April 17, 2014

+ Research questions 4 n RQ1. How can the attitude of the participants in user forums be made explicit and recorded within the structure of a discussion? explicitly represented? n RQ2. How can a structured discussion support decisions on ✓ what comments should be further analyzed? ✓ Argumentation-based Discussion for User Forum: A Research Preview Itzel Morales Ramírez Thursday, April 17, 2014

+ Principal Ideas 5 n Argumentation is a natural way humans engage in a discussion, based on arguments, in order to clarify or defend their opinions. n Represent an online discussion as a structured set of arguments according to the Argumentation Framework (AF). n Dung’s framework: y 1 n An argumentation framework (AF) is a pair: attacks n AF= <A, Def> n A is a set of arguments. A={y 1 , … ,y n } attacks (an argument is an abstract entity) n Def is a binary relation of defeat between arguments . y 3 y 2 attacks n Def ( y i ,y j ) means that y i represents an attack against y j. n Ex1. A={y 1 ,y 2 ,y 3 } Def={(y 1 ,y 3 ), (y 3 ,y 1 ), (y 2 ,y 3 )} Argumentation-based Discussion for User Forum: A Research Preview Itzel Morales Ramírez Thursday, April 17, 2014

+ Recalling RQ1: 6 Explicit participant’s attitude n Computational construction of the argumentation-based discussion forum (ADF) as a directed acyclic graph G=(V,E) Real number representing Comment know c participant’s perception on ID Participant their own expertise level. n (V)ertex Description Participant’s attitude supports rejects neutral n (E)dges Argumentation-based Discussion for User Forum: A Research Preview Itzel Morales Ramírez Thursday, April 17, 2014

+ Example 7 Excerpt of the discussion #112163 in AOOo bugzilla c0 1 0 Dotan Cohen supports Calc lets users rejects have multiple … c1 8 michael.r... 0.75 1 eric.savary 0.5 rejects Reassigned to duplicate requirements. c2 3 Dotan Cohen 1 neutral Thanks, ES, but supports c3 this is not … 2 eric.savary 0.5 11 timdeaton 0.5 closed c8 Dotan’s … explain usefulness of this c11 ADF graph Argumentation-based Discussion for User Forum: A Research Preview Itzel Morales Ramírez Thursday, April 17, 2014

+ Recalling RQ2: 8 Computing Relevant Comments n Algorithm for a quantitative evaluation based on graph search algorithms, it evaluates the entire discussion. 1 0 Dotan Cohen supports Calc lets users rejects have multiple … Discussion 8 michael.r... 0.75 1 eric.savary #112163 0.5 rejects Reassigned to duplicate requirements. Total comments 3 Dotan Cohen 1 12 neutral Thanks, ES, but supports this is not … 2 eric.savary 0.5 11 timdeaton 0.5 r closed s Dotan’s … explain 4 usefulness of this n r 6 5 s s 7 9 10 Argumentation-based Discussion for User Forum: A Research Preview Itzel Morales Ramírez Thursday, April 17, 2014

+ Research plan 9 n Experiments using a graph library to manipulate graphs (with extensions) ADF Graph Output Input (txt files) y 0 Discussions rejects #9437 (47 comments) *4 more to be analysed y 2 y 1 supports n Investigating Neo4j (for manipulating and storing huge data) hasCustomers Argumentation-based Discussion for User Forum: A Research Preview Itzel Morales Ramírez Thursday, April 17, 2014

+ Conclusion 10 n Present the Argumentation-based Discussion Forum (ADF) n Make an explicit representation of participants’ attitude (“in favour” or “against”) n Algorithm that will allow participants to get an overview of the discussion and point out relevant comments n Research plan Argumentation-based Discussion for User Forum: A Research Preview Itzel Morales Ramírez Thursday, April 17, 2014

+ Thank you very much for your 11 attention! Any questions? Argumentation-based Discussion for User Forum: A Research Preview Itzel Morales Ramírez Thursday, April 17, 2014

+ Recalling RQ2: 12 Computing Relevant Comments n Algorithm for a quantitative evaluation based on graph search algorithms, it evaluates the entire discussion. n Definition 5. Support relation strength. A given comment y j has associated a support strength that is calculated based on the number of incoming support relations and the parameter know c. y 0 n Definition 6. Rejection relation strength. Idem supports rejects supports y 3 know c y 1 y 2 n Definition 7. Relevant comments (RelC). A set of comments where given a comment y belonging to the set there are no y 0 comments, included in the set, that reject y . supports supports y 3 y 2 Argumentation-based Discussion for User Forum: A Research Preview Itzel Morales Ramírez Thursday, April 17, 2014

+ Overall approach 13 Systematic approach for acquiring end-user feedback and deriving requirements knowledge from it . Argumentation-based Discussion for User Forum: A Research Preview Itzel Morales Ramírez Thursday, April 17, 2014

+ Parallel ongoing work 14 n In parallel with this work, we are performing research on how to use NLP to identify the attitude in unstructured text. n Attitude that is derived from patterns in a conversation or that can be derived from the intention of rejecting or supporting statements in a conversation. < performance of Action < performance of * Contribution < performance * 1 Perception inheres in > Agent of * *triggered by > * 1..* Communicative Act Intentional < propositional Moment * 3 content of 1 1 involves > Sender Receiver 1 Speech act Belief Desire Intention Topic Proposition (internal commitment) < inheres in 1 propositional 1 Message Communicative Reflexive content of > Intention Intention Argumentation-based Discussion for User Forum: A Research Preview Itzel Morales Ramírez Thursday, April 17, 2014

+ Extension to the 15 Communication ontology (CONTO) Non < is interpreted in Speech act 1 linguistic 1 terms of act {disjoint} Gesture Drawing Constantives Directives Attached information Expressives Assertives Requestives Questions Confirmatives Informatives Concessives Requirements Responsives Descriptives Suggestives Thank Reject Accept Suppositives Negative Positive opinion opinion Argumentation-based Discussion for User Forum: A Research Preview Itzel Morales Ramírez Thursday, April 17, 2014

+ What is explicit user feedback? 16 From your smile I I’m very guess you happy Explicit Implicit are happy today User feedback Meaningful information given by users of software applications with the purpose of suggesting new needs or improvements to them. Argumentation-based Discussion for User Forum: A Research Preview Itzel Morales Thursday, April 17, 2014 Ramírez

+ Implementation (mock-up) (1) 17 Argumentation-based Discussion for User Forum: A Research Preview Itzel Morales Thursday, April 17, 2014 Ramírez

+ Implementation (mock-up) (2) 18 Argumentation-based Discussion for User Forum: A Research Preview Itzel Morales Thursday, April 17, 2014 Ramírez

+ Implementation 19 n Coordination with a colleague in Mexico: Eloy Fragoso (flea07mx@gmail.com) Argumentation-based Discussion for User Forum: A Research Preview Itzel Morales Ramírez Thursday, April 17, 2014

+ Knowledge level classification 20 [Ericsson2006] n A novice is a person who will be considerably overloaded in seeking to gather information, understand what it means. n An initiate is a person who has at least an introductory knowledge of the domain. n An apprentice is a person who is learning beyond the introductory knowledge of the domain. n An advanced is a person who has acquired certain knowledge that let her to perform activities unsupervised. n An expert is someone who exhibits a high degree of competence. Argumentation-based Discussion for User Forum: A Research Preview Itzel Morales Thursday, April 17, 2014 Ramírez