Presentation of ArgMAS2010.

1. Context Proposal Conclusions Refs. Questions
Towards a dialectical approach for
conversational agents in selling situations
Maxime MORGE - Sameh ABDELNABY - Bruno BEAUFILS
May 2010
Morge, Abdel-Naby & Beaufils ArgMAS 2010 - Page 1

2. Context Proposal Conclusions Refs. Questions
Industrial context: commercial trading website
§ Online presence of companies is
+ cost-efficient
- lacking of a persuading salesman (low transformation ratio)
Morge, Abdel-Naby & Beaufils ArgMAS 2010 - Page 2

3. Context Proposal Conclusions Refs. Questions
CSO Language Processor [Roberts & G¨lsdorff’07]
u
§ Manage sessions
§ Handle misspellings
§ Language dependent preprocessing
§ Select and carry out best system answer
according to a recognition pattern.
§ Interact with back end (e.g. databases)
§ ...
§ From reactive agents to sales-driven dialogues
Morge, Abdel-Naby & Beaufils ArgMAS 2010 - Page 3

4. Context Proposal Conclusions Refs. Questions
Dialogue: Phases & Purposes
§ BEFORE-SALE: 2 processes interleaved
§ needs identification, i.e. information-seeking to share the user
requirements
§ product selection, i.e. information-seeking to select relevant
products
§ SALE: negotiation (offer/accept/reject)
§ single sale, i.e. user/agent rank features of the products
§ cross-selling, .i.e. additional products
§ up-selling, i.e. similar products
§ AFTER-SALE: enquiry/deliberation for problem detection and/or
product replacement.
Initial situation → Conflict Open problem Ignorance of
Goal ↓ a participant
Stable agreement persuasion enquiry information
i.e., Resolution seeking
Practical settlement negotiation deliberation none
i.e., Decision
[Walton’95]
Morge, Abdel-Naby & Beaufils ArgMAS 2010 - Page 4

5. Context Proposal Conclusions Refs. Questions
Dialogue-game protocol for information-seeking/negotiation
request introduce
question
accept reject
assert unknow
Morge, Abdel-Naby & Beaufils ArgMAS 2010 - Page 5

6. Context Proposal Conclusions Refs. Questions
Virtual Seller Agent Vs. User: Before-Sale & Sale scenarios
1 VSA: Your moves through our website indicates that you might be
interested in Quilts, do you need help to select one?
2 User: Nice.
3 VSA: Are you allergic?
4 User: Yes.
5 VSA: Do you sweat?
6 User: Sure. Is there any organic quilts ?
7 VSA: Definitely, but which dimensions are you looking for?
8 User: 200x200. Is there any one within a budget of e 350?
9 VSA: Yes, it cost e 239 since the dimensions are 200x200, it is
nonallergenic and organic.
10 User: I buy it.
11 VSA: Since it is still within your budget, I may also suggest you to
buy a “four seasons” one that cost e 279?
12 User: Good idea.
13 VSA: What about a quilt cover that cost e 99 only? still in your
budget.
14 User: No, thanks.
Morge, Abdel-Naby & Beaufils ArgMAS 2010 - Page 6

7. Context Proposal Conclusions Refs. Questions
Scenario Formalization
1 VSA: question(is(product,quilt) because search(you,product).
2 User: assert(is(product,quilt)).
3 VSA: question(is(you,allergic)).
4 User: assert(is(you,allergic)).
5 VSA: question(is(you,sweat)).
6 User: assert(is(you,sweat)). question(is(product,organic)).
7 VSA: assert(is(product,organic)). question(dimension(product,200,200)).
8 User: assert(dimension(product,200,200)). question(budget(product,350)).
9 VSA: introduce(is(product,quilt) because budget(product,239) and is(product,
nonallergenic)) and is(product,organic)).
10 User: accept(is(product,quilt).
11 VSA: introduce(is(product,quilt) because budget(product,279) and
is(product,nonallergenic)) and is(product,fourseasons)).
12 User: accept(is(product,quilt).
13 VSA: introduce(is(product, quilt)) and is(product,quiltcovers) because budget(product,
333.90) and is(product, nonallergenic)) and is(product,fourseasons)).
14 User: reject(is(product, quiltcovers).
Morge, Abdel-Naby & Beaufils ArgMAS 2010 - Page 7

8. Context Proposal Conclusions Refs. Questions
Assumption-based argumentation for dialogue [Morge & Mancarella ArgMAS’07]
L=G∪D∪B
r11 : infoseeking(user) ← send(question(is(user, allergic))),
∼ is(user, allergic), is(product, quilt), ∼ receive(x)
r12 : infoseeking(user) ← send(question(is(user, sweat))),
∼ is(user, sweat), is(product, quilt), ∼ receive(x)
r21 : infoseeking(product) ← send(question(is(product, quilt))),
search(user, quilt), ∼ is(product, quilt)
r22 : infoseeking(product) ← send(question(is(product, nonallergenic))),
∼ is(product, nonallergenic), ∼ receive(x)
...
r29 : negotiating(product) ← send(introduce(product)), budget(product, y )
r31 : budget(product, 350) ← is(product, nonallergenic),
is(product, organic), dimension(product, 200, 200)
r33 : is(product, organic) ← is(user, sweat)
Asm = {send(x), ∼ receive(x)}
Con(send(assert(x))) = {send(assert(y )), send(question(z)), . . .}
Con(send(question(x))) = {send(assert(z)), send(question(y )), . . .}
...
infoseeking(user)Pinfoseeking(product)
infoseeking(product)Pnegotiating(product).
Morge, Abdel-Naby & Beaufils ArgMAS 2010 - Page 8

9. Context Proposal Conclusions Refs. Questions
Argument: deduction tree [Morge & Mancarella ArgMAS’09]
Attacks comes from the contraries:
§ The VSA asks if the user is allergic;
§ The VSA asks if the product must be nonallergenic.
infoseeking(user) infoseeking(product)
question(is(user, allergic)) question(is(product, nonallergenic))
∼ is(product, nonallergenic) ∼ is(product, nonallergenic)
Morge, Abdel-Naby & Beaufils ArgMAS 2010 - Page 9

10. Context Proposal Conclusions Refs. Questions
To take away: conversational agents in selling situation
§ Approach:
§ A proactive agent which can query the user.
§ A selling agent which can justify its proposals.
§ Technologies:
§ www.MARGO.sf.net handles dialogue strategy to generate and
evaluate utterances
§ CSO LP handles natural language processing
HTTP request
Web server (apache)
HTTP reply
request
UI NL NL
request reply
reply (browser)
VSA (LIFL/AS) ACL request
MARGO CSO LP
ACL reply
§ In progress: development, implementation, evaluation and validation
Morge, Abdel-Naby & Beaufils ArgMAS 2010 - Page 10

11. Context Proposal Conclusions Refs. Questions
References
J. A. George Ferguson. Mixed-initiative systems for collaborative problem solving.
AI Magazine, 28(2):23–32, 2007.
C. Rich, C. L. Sidner, and N. Lesh. COLLAGEN applying collaborative discourse theory to
human-computer interaction.
AI Magazine, 22(4):15–25, 2001.
D. Sadek. Multi-Agent Programming, chapter Artimis Rational Dialogue Agent Technology:
An Overview, pages 217–225.
Springer-Verlag, 2005.
D. Walton and E. Krabbe. Commitment in Dialogue.
SUNY Press, 1995.
Roberts, F., G¨lsdorff, B.: Techniques of dialogue simulation. In: Proc of the 7th
u
International Conference on Intelligent Virtual Agents. Volume 4722 of Lecture Note in
Computer Science., Paris, France, Springer (2007) 420–421.
M. Morge and P. Mancarella. The hedgehog and the fox. An argumentation-based decision
support system.
In Proc. of the 4th International Workshop on Argumentation in Multi-Agent Systems
(ArgMAS), pages 1–18, Honolulu, Hawai, 2007.
M. Morge and P. Mancarella. Assumption-based argumentation for the minimal concession
strategy.
In Proc. of the 6th International Workshop on Argumentation in Multi-Agent Systems
(ArgMAS), pages 1–18, Budapest, Hungary, 2009.
Morge, Abdel-Naby & Beaufils ArgMAS 2010 - Page 11

12. Context Proposal Conclusions Refs. Questions
Mixed-initiative dialogue systems for collaborative problem solving
§ Existing dialogue systems:
§ TRAINS-93 [Ferguson 07]
§ Collagen [Rich et al. 01]
§ Artemis Agent Technology [Sadek 05]
§ “I want to purchase a quilt”
1 It can be a direct report of a need;
2 It can be a statement of a goal that a user is pursuing independently;
3 It can be a proposal to adopt this joint goal.
§ Complexity and heuristics of goal recognition
§ But small is beautiful . . .
§ Dialectical approach
Morge, Abdel-Naby & Beaufils ArgMAS 2010 - Page 12