TW Wine Agent

From Tetherless World Wiki

TW Wine Agent
type	Research Project
status	active
homepage	http://onto.rpi.edu/wiki/wine
Management
team	Tetherless World Constellation
participant	James Michaelis,Deborah L. McGuinness,Li Ding
Tags
tag	Explanation,Agent,Trust,Social Semantic Web
relation	Inference Web,Semantic Portal Wiki
Internal

Contents 1 Overview 2 Design of Program 2.1 Components/Tasks 2.2 Implementation Schedule 3 Additional Design Notes 4 June 11 Presentation 5 KSL wine agent 6 resources

Overview

The KSL Wine Agent demonstrates a functional example of an ontology-powered web service providing expert advice using web tools including an online reasoner (JTP), query engine (OWL-QL), and explanation environment (Inference Web). Early versions of the Wine Agent relied solely upon information contained in predefined food and wineontologies, local wine ontologies, and online wine directories, it used one set of predetermined rules for pairing wine and food combinations. While this approach provides a good demonstration of ontology usage and explanation, it does not account for potentially multiple different user opinions on ideal food-wine pairings. In this paper, we describe a new version of the Wine Agent, which: (1) makes wine recommendations based on multiple user opinions, (2) introduces mechanisms for resolving conflicts between user wine recommendations, and (3) provides initial explanation services for supporting its recommendations.

Design of Program

Components/Tasks

NOTE: Each programming task listed below is assigned a corresponding reference number.

The TW Wine Agent will allow users to both add and view wine recommendations for a given set of foods.

ADDING RECOMMENDATIONS: This is handled through the Wine Agent Wiki.

(13) Taking recommendation input from users - DONE
(14) Encoding Food Ontology into wiki - DONE
(15) Encoding Wine Ontology into wiki - IN PROGRESS

VIEWING RECOMMENDATIONS: This is handled directly on the Wine Agent. The process involves the program handling the following tasks:

- Displaying a food taxonomy and taking user input - Querying the Wine Agent Wiki for relevant recommendations - Displaying a detailed breakdown of these recommendations, along with their associated wines - Allowing for filtering of results - Handling Inference Web based proofs for displayed results

The relevant subtasks for each of these are given below:

VIEWING FOOD TAXONOMY:
(1) Displaying a taxonomy of foods derived from the food ontology - DONE
(2) Taking a selection from this taxonomy as input - DONE
(3) Displaying recommendation counts (from the wiki) for each food in the taxonomy, as well as a separate count of the recommendations for all subclasses - DONE

QUERYING THE WINE AGENT WIKI:
(4) Retrieving recommendations explicitly for the selected food (e.g., for a Bland Fish query, retrieving the Bland Fish recommendations) - DONE
(5) Retrieving recommendations for subclasses of the selected food (for a Bland Fish query, retrieving recommendations for Halibut, Scrod, and Flounder) - IN PROGRESS
(6) Retrieving recommendations for superclasses of the selected food (for a Bland Fish query, retrieving recommendations for Fish and Seafood) - IN PROGRESS
(7) In case of encountering two or more recommendations from the same author, determining that the recommendation for the subclass or superclass food should be disregarded - IN PROGRESS

VIEWING RECOMMENDATIONS:
(8) Displaying a user-by-user breakdown of each relevant recommendation - DONE
(9) Displaying a list of associated wines for the relevant recommendations, as well as numbers of relevant recommendations - DONE

FILTERING RECOMMENDATIONS:
(10) Implementing recommendation filters for the following recommendation properties:
a) Food recommendation made for - IN PROGRESS
b) Recommender properties (expert vs. novice, and so on) - IN PROGRESS
c) Wine properties (color, body, etc.) - IN PROGRESS
NOTE: Current plan is to finish implementing these filters in order of presentation.

INFERENCE WEB PROOFS:
(11) Integrating proof links ("why" button) in appropriate places - IN PROGRESS
(12) Displaying pop-ups with contained inference web based proofs - IN PROGRESS

Implementation Schedule

Currently planning to complete the remaining tasks according to the following schedule:

Week of 5/25: TASKS 5,6,7,10a,10b,10c
Week of 6/1: TASKS 15, 11, 12*

NOTE: For task 12, James would like to meet with Deborah and/or Li to determine best way to implement this.

Additional Design Notes

Navigation Issues: At present, the Wine Agent relies upon a series of browser tokens which denote different sections of the application. These are defined as follows:

ACKROOT: (Acknowledgement Root Panel) Acknowledgements screen
OVRROOT: (Overview Root Panel) Wine Agent Overview screen
INTROOT: (Intro Root Panel) Introductory screen for Wine Agent
RCMROOT: (Recommendation Making Root Panel) adding screen (displays link to Wine Agent Wiki)
RCFROOT: (Recommendation Food Root Panel) Food taxonomy display (allows user to select food for viewing recommendations on)
RCBROOT: (Recommendation Browsing Root Panel) Displays recommendation details for selected food, as well as accompanying wines

In this terminology, a Root Panel contains all the contents for a given section of the Wine Agent.
Currently, only the INTROOT panel can be referenced externally. However, I am currently in the process of revising the Wine Agent hyperlink management system to allow for external references to other pages.

Usage on mobile devices (iPhone, etc.): Currently, the TW Wine Agent seems to function normally on the iPod Touch. However, since the TW Wine Agent was designed for non-mobile screen resolutions (1024*768 and up), it becomes necessary to zoom in to read text and click links (as with any standard webpage). If time permits, perhaps we could look into designing an iPhone specific interface for the TW Wine Agent. Regardless, if we want to show something on the iPhone for June 11th, it should be doable.

June 11 Presentation

I think the storyboard content from our AAAI proposal (http://tw.rpi.edu/internal/wiki/images/b/b6/ISD-AppMichaelisJ.pdf) would supplement this presentation well, in the form of demos to supplement presentation content. As I recall, the AAAI reviewers felt this storyboard was well thought out (but could have used more explanation of AI topics).

However, since a number of changes have occurred in the Wine Agent design since this original storyboard (implementing the Wine Agent Wiki, and so on), some changes will need to be made. I have thought up 5 possible demos based around the AAAI storyboard, which I am trying to integrate into the presentation outline below:

INTRODUCTION:
- Briefly introduce the concept of social semantic web
- Address some of the problems that it could resolve
- Introduce the Wine Agent concept as a test application for social semantic web

THE KSL WINE AGENT:
- State some background on the KSL Wine Agent (See overview section of wiki page)
- SHOW KSL WINE AGENT (possibly?)
- Address how KSL Wine Agent lacks ability to account for variable user opinion, or derive input from users

THE TW WINE AGENT:
- Introduce the TW Wine Agent as an extension of the KSL Wine Agent for handling social aspect of semantic web
- Give a brief outline of the differences between the KSL and TW Wine Agents, highlighting the introduction of the semantic wiki

USE OF WIKI:
- Encodes data in RDFS
- Allows for users to easily define wine recommendations for different kinds of food
- PRESENT DEMO 1
- Can also allow for users to define new kinds of food, and new kinds of wine
- However, expressivity of RDFS limits the possible relationships that can be encoded in the semantic wiki data
- Highlight how the food and wine ontologies can supplement the semantic wiki with additional DL based encodings for data

SOCIAL ASPECTS OF WINE AGENT:
- Highlight ability for anyone to contribute to the TW Wine Agent knowledge base
- PRESENT DEMO 2
- Also, highlight the ability for users to selectively filter recommendations
- PRESENT DEMO 3

INFERENCE WEB USAGE:
- PRESENT DEMO 5

CONCLUDING REMARKS

DEMO 1: A user wants to enter a wine recommendation for Swordfish. (This is based off TASK 1 from the AAAI storyboard)

For this, we can go through the recommendation adding sequence used on the semantic wiki. At this point, we can provide some explanation of how the wine agent is able to extract encoded descriptions of wine recommendations from this specialized wiki.

DEMO 2: A user decides to query for wine recommendations for Steak. (This is based off TASK 2 from the AAAI storyboard)

As in the original slide, the user will be selecting the food Steak from the food taxonomy we have implemented. This will result in the user being taken to a screen with two tabs (one for viewing each applicable recommendation, and one for viewing the wines obtained through these recommendations, along with total number of recommendations for each wine).

DEMO 3: The user decides to experiment with the recommendation filters (This is based off TASK 3 from the AAAI storyboard)

I think for this we should provide filtering options for one or more of the following categories for this demonstration:

1) Food from recommendation
2) Recommender properties (expert vs. novice, and so on)
3) Wine properties (color, body, etc.)

I am currently planning on trying to get each of these filters working in the order shown above. I feel that each of these three filters will allow for a good demonstration of user-oriented data narrowing, as well as SPARQL based querying.
DEMO 4: Observing conflicts. - Earlier we spoke about highlighting recommendation conflicts similar to the one given in TASK 5 of the AAAI storyboard. Would we still like to do something like this? More details here later.

DEMO 5: Inference Web Proofs. - Display a popup containing a proof for some given data on the Wine Agent. Details to come later.

KSL wine agent

• http://www.ksl.stanford.edu/people/dlm/webont/wineAgent/

• Citation: Eric I. Hsu and Deborah L. McGuinness. (2003) Wine Agent: Semantic Web Testbed Application. In Proceedings of the Workshop on Description Logics, 2003.

Publication inproceedings ( Edit )
type	InProceedings
bibtype	inproceedings
Bibtex basics
author	Eric I. Hsu and Deborah L. McGuinness
title	Wine Agent: Semantic Web Testbed Application
booktitle	Proceedings of the Workshop on Description Logics
address	Rome, Italy
year	2003
Bibtex more
Access Paper
abstract	The Wine Agent is a demonstration system that uses an underlying domain ontology to provide suitable wines for a given meal. In doing so it serves as a testbed, not only for the logical domain description, but additionally for emerging Semantic Web technologies that process, infer, justify, and execute the pairings. Specifically, it combines the DAML+OIL and OWL Web-based description logics with the JTP theorem prover. The resulting knowledge base can be queried remotely via the DQL query language. Suitable pairings are explained within the Inference Web apparatus, and then transacted via a preliminary implementation of Web Services. Besides serving as a prototype for these methodologies, the wine agent has provided useful empirical lessons regarding reasoning via Semantic Web axioms, language requirements, and requirements for explanation, as well as pragmatic issues concerning implementation and integration.

KSL Technical Report ID: KSL-03-17

resources

• tw:WineAgentResource