Sara von Mosch, (2005) ConversationLens: Towards Personalized Online Conversation

---

Comments by ElizabethWindram

ConversationLens sounds like a great idea and an interesting approach to resolving several issues that currently exist in threaded conversation (especially similar/identical threads that exist in multiple places).

I am a little unclear though on where a person begins. I logged into MovieLens and rated enough movies to get going with the system. I am able to do a number of things, but don’t seem to be connected to any type of threaded discussion…

In your figures, it does appear there are conversations that I can view and partake in. This seems extremely valuable, especially if the system is plucking all related comments and posts pertaining to the subject at hand. But do I begin with a search? Are the conversation entry points available for me all the time while I am browsing? I would assume if I were interested in a movie and MovieLens suggested that I would probably rate this title highly, I’d like to hear what other people had to say about the movie. Which I would further assume is where the conversation threads come in? Is this true?

This really sounds like a great idea and a wonderful front-end, personalized entry into previously difficult-to-wade-through threaded conversations.

ElizabethWindram

---

Comments by CliffLampe

I'm really intrigued by the attempt to apply MovieLens architecture to asynchronous conversation. I don't think I'm quite getting all the ramifications of the "entity" model, but I think they are objects that can be rated. In that case, other entities are possible besides what's listed in the paper. For example, threads exist on many levels, and you might treat a sub-thread that's really interesting as another entity separate from an interesting forum. Some neat possibilities.

You mention Slashdot, so I'm constitutionally unable to not comment on your point there. You say: "Slashdot, for example, lets users adjust the level of filtering based on the scores given by community moderators. While this technique, is an effective noise filter, it shows content the moderators, not the individual viewing the material, have rated highly. " However, Slashdot does allow for individual customization of message rating in personal user profiles. Users can tweak ratings based on a couple dozen factors, including who posted the comment, what type of comment it was and so forth. This isn't really dynamic changes based on their rating behavior, so isn't a huge factor for ConversationLens, but still should be clear when discussing Slashdot. The factor in this customization that I do think applies to your work is the significant number of people who change their permanent filter settings to read at the lowest level. When some people seem to want the information overload scenario, what implication does that have in ConversationLens?

One could easily imagine adding some dynamic recommendation based on user behavior in Slashdot or other rating-based discussion forums. If you open the moderation to more people, for example, you could use how they rate comments as a basis or recommending other comments to them. That is to say, if I moderate mostly "Funny" I might want some differential applied to comments others also label as funny.

Another (fun?) thing to consider with ConversationLens is how to abuse it. I can imagine some trolls coming in, posting a pretty bad comment, and then using the rating systems and multiple accounts to get that pretty widely spread in the system. It'd be interesting to see how much gaming occurs on MovieLens, though my intuition is not much. Online conversations seem to be different that way, as evinced by interactions on Usenet and Slashdot. I'm really interested to hear more about the nuts and bolts of this system.

CliffLampe

---

Comments by AdelePontes

As CliffLampe, I'm also unsure about how well the MovieLens architecture may be applied to asynchronous conversation. My main concern regards the notion of *context* in conversation.

One of our research issues is to attempt to characterize the role of the context of production in helping the interpretation of the communicative exchanges. When you say "this system allows for content to be displayed when it is relevant and of interest to the user, regardless of the context in which it was created", I wonder what is your definition of context here, and whether the context of production is considered by ConversationLens during content selection, content presentation, or at all.

Have you investigated the applicability of the ConversationLens to domains that are unstructured or don't have a structured referent (such as the "Movie" entity)?

AdelePontes

[Please insert additional comments in this section followed by your WikiName]