Providing Expert Advice by Analogy for On-Line Help

Providing Expert Advice by Analogy for On-Line Help Henry Lieberman and Ashwani Kumar Media Laboratory Massachusetts Institute of Technology Cambridge, MA, USA http: //www. media. mit. edu/~lieber/ Common Sense Computing • MIT Media Lab

What happens when interaction with the Web goes wrong? Increasingly, people use the Web to perform procedures (buy things, vote, date, seek jobs) as well as just browse What happens when things go wrong? Debug it yourself Get help Increasingly, confidence in Web interactions depends on effectiveness of help Common Sense Computing • MIT Media Lab

How do you get help? Look up things in problem-solution database Telephone support On-line chat Common Sense Computing • MIT Media Lab

Elicititation & Explanation Elicitation: Helper asks user what is wrong. Get enough information from user to determine problem and choose solution. Explanation: Helper tells user how to fix the problem. Explain why the solution worked, and how to avoid such problems in the future. Common Sense Computing • MIT Media Lab

Mismatch between expert and novice models The helper is an expert; the user is a novice Novice may lack technical vocabulary to understand the elicitation questions Novice may lack background knowledge to understand explanation of soultion Expert may be unable to empathize with novice Common Sense Computing • MIT Media Lab

Problem/Solution Database Common Sense Computing • MIT Media Lab

AI Expert Systems are the traditional way to encode expert behavior Knowledge Engineer experts to get their knowledge and procedures Encode Expert Model Deep but narrow knowledge If expert and novice don’t share vocabulary and knowledge, it’s difficult for the novice to interact directly with the Expert Model Common Sense Computing • MIT Media Lab

We need a model of novice knowledge How do we model the knowledge of someone who is not supposed to know very much? Traditionally, novice is modeled as subset of expert knowledge Intelligent Tutoring Systems, Qualitative Physics But how do you model novice knowledge “in general”? Common Sense Computing • MIT Media Lab

Open Mind - Push Singh Common Sense statements collected from volunteer contributors on the Web 770, 000 English sentences Conventional data/knowledge bases: Do a lot of organizing work up front, so you can get the stuff out easily OM: Get the stuff in any which way, then do a lot of work on the back end to interpret it Query expansion, Parser, Semantic net miner Common Sense Computing • MIT Media Lab

Open Mind - Push Singh Common Sense Computing • MIT Media Lab

How do human helpers succeed in bridging Expert/Novice gap? We interviewed helpers on the America Online (internet service provider) help desk Helpers tend to provide “cookbook” solutions We asked for successful interactions where users came to understand the problem and solution Helpers best explained things by making analogies to everyday life situations! Common Sense Computing • MIT Media Lab

Problem/Solution Database Problem: “My browser can see my company’s home page, but it won’t let me access my workgroup’s Web site” Solution Procedure: “Check if cipher strength is '0’. Upgrade Browser to 128 bit Encryption. ” Explanation: Generally, Websites require 128 -bit encryption in order to process information securely. If the cipher strength of your browser is inadequate, you will not get into secure Websites. Upgrading your browser's encryption may help it better handle secure Websites. NOTE: You only need to do this when unable to get to secure Websites. Common Sense Computing • MIT Media Lab

Better: Explain by Analogy Encryption in a browser is like security clearance to enter a building. If you don't have the proper security clearance, you may be able to get into the building, but not into certain areas. You must upgrade your security clearance status to go further. So without the proper encryption, your browser may be able to access a website, but not log in. Common Sense Computing • MIT Media Lab

Introducing Suggest. Desk is an agent that “listens” to the helper/user chat Provides suggestions to helper (unseen to novice user) Tries to recognize opportunities for making analogies between problem/solution DB and Common Sense knowledge Common Sense Computing • MIT Media Lab

Suggest. Desk Common Sense Computing • MIT Media Lab

Example User: “My browser runs slowly” Problem/Solution DB: Browser might run slowly because of network congestion Browser might run slowly because the browser is infected by a virus Common Sense Computing • MIT Media Lab

Elicitation & Explanation Elicitation: (Traffic) When did you try to log in? (Virus) Have you downloaded any new applications lately? Solution: (Traffic) Try again at a later time. (Virus) Run an anti-virus program. Common Sense Computing • MIT Media Lab

Suggest. Desk’s help (Traffic) What makes things slow down? Traffic runs slowly at rush hour. Analogy: The AOL service is like a road. The users are like cars. If too many users try to use the service at the same time, it slows down. The solution is to try at a time when things are less crowded or find an alternative route. Common Sense Computing • MIT Media Lab

Suggest. Desk’s help (Virus) What makes things slow down? People slow down when they are tired. Being sick can make you tired. Analogy: A computer virus is like a biological virus If you have the flu, you can’t do things as fast as you normally would An antivirus program is like medicine. Common Sense Computing • MIT Media Lab

Suggest. Desk Architecture The Natural Language Processor (NLP), The Commonsense Processor (CP), The Expert Analyzer (EA), The Analogy Mapping Engine (AME), and The Elicitation and Explanation Processor (EEP). Common Sense Computing • MIT Media Lab

$Natural Language Processing Via Monty. Lingua, Common Sense part-of-speech tagger Result= [{prep_phrases_tagged=[], verb_phrases_tagged=[is/VBZ running/VBG],$

Natural Language Processing Via Monty. Lingua, Common Sense part-of-speech tagger Result= [{prep_phrases_tagged=[], verb_phrases_tagged=[is/VBZ running/VBG], verb_arg_structures_concise=[("run" "browser" "slow")], noun_phrases=[browser], noun_phrases_tagged=[browser/NN], adj_phrases_tagged=[slow/JJ], verb_arg_structures=[[is/VBZ running/VBG, browser/NN, [slow/JJ]]], modifiers_tagged=[slow/JJ], prep_phrases=[], verb_phrases=[is running], parameterized_predicates=[[[run, [past_tense, passive_voice]], [browser, []], [slow, []]]], modifiers=[slow], adj_phrases=[slow]}] Common Sense Computing • MIT Media Lab

Expert. Net (Effect. Of 'surf internet' 'download files') (Effect. Of 'surf internet' 'download applications') (Effect. Of 'download files' 'browser cache is large') (Effect. Of 'download applications' 'browser infected by virus') (Effect. Of 'PC infected by virus' 'browser run slow') Common Sense Computing • MIT Media Lab

Analogy Mapping Engine Analogies: [[computer, [[Used. For, surf internet, 1. 1887218755408673], [Capable. Of. Receiving. Action, run slow, 1. 1887218755408673], [Capable. Of. Receiving. Action, crash, 1. 1887218755408673], [Capable. Of. Receiving. Action, start, 1. 1887218755408673]], 6. 1887218755408675], [car, [[Capable. Of. Receiving. Action, damage, 1. 1887218755408673], [Capable. Of. Receiving. Action, crash, 1. 1887218755408673], [Capable. Of. Receiving. Action, start, 1. 1887218755408673]], 5. 930167946706389], [software, [[Capable. Of. Receiving. Action, run slow, 1. 1887218755408673], [Capable. Of. Receiving. Action, crash, 1. 1887218755408673], [Capable. Of. Receiving. Action, install, 1. 1887218755408673]], 5. 855516191543203]] Common Sense Computing • MIT Media Lab

Weighting log(f+0. 5*i+4), where f=outgoing edges and i = incoming edges Common Sense Computing • MIT Media Lab

Suggest. Desk Architecture The Natural Language Processor (NLP), The Commonsense Processor (CP), The Expert Analyzer (EA), The Analogy Mapping Engine (AME), and The Elicitation and Explanation Processor (EEP). Common Sense Computing • MIT Media Lab

User Testing Common Sense Computing • MIT Media Lab

Woodstein (with Earl Wagner) Debugger for Web procedures By analogy to debugger for programming Provides visualization, explanation, tools for incremental exploration Self-help debugging Co-operative debugging between expert and novice Common Sense Computing • MIT Media Lab

Woodstein Common Sense Computing • MIT Media Lab

Conclusion The effectiveness of online help is key to the success of Web interactions Help needs to bridge the gap between expert and novice knowledge Common Sense Reasoning can help find analogies that allow expert and novice to communicate We need debuggers that help us systematically explore the causes of problems Don’t worry, help is on the way! Common Sense Computing • MIT Media Lab