user story: UI for topic classifying items

(by Bruce Smith -- feel free to embed questions or enhancements if you sign them; for major revisions or new topics, a new page would be better)

==

This is a kind of "user story" (or a few of them) meant to try to be less vague about how the UI (of a software tool supporting OPSN -- see blog post) might work for topic-classifying items (posts and comments).

==

background:

Someone makes a blog post about how it's important to reduce human CO2 output, but difficult. They refer to a famous article about a "list of wedges", each of which is hard to do, and we need to do 13 of them to reduce the output enough.

There are various replies to this post (as comments), which I will summarize or paraphrase as single phrases (since understanding exactly what they say or mean is not important for this example):

• do we really need to reduce human CO2 output?

• wouldn't thorium reactors solve the problem?

• use biochar as a solution

• use accelerated weathering (olivine)

• extract uranium from seawater

  • there is a working demo system for this

==

As a first step to help organize the discussion, some people want to classify the post and comments by topic.

For example, they want to mark the comment about thorium reactors with the topic phrase "thorium reactors", and perhaps also disambiguate that to associate it with a more specific meaning of that phrase.

The writer or any reader can do that.

(By default, until a conflict arises, the semantics of multiple people being able to do it is "wikilike", meaning anyone can do it and everyone sees the result.)

(We'll call whoever does one act of adding a topic to an item "the user", in this story.)

We assume they (the user) are viewing the items (posts and comments) in a special UI using a special software tool ("the system"), and that it has access to internal features similar to "google search".

The system can automatically find good candidates for topic phrases -- sequences of a few words which are more common in these items, and the past items seen by this user, than in random text, and which are present in some items but not others.

When the user wants to classify items, he or she gets into a special UI mode in which candidate topic phrases are highlighted.

Mousing over one, we see some info in a panel to the right, the "phrase info panel", which indicates how common it is in three sets of items (the present few we're seeing, a larger set we know about (defined by our user prefs), and all known text), and a list of the few most likely topic names it might refer to (disambiguated as needed -- internally like urls or exprs, but shown to the user as nicknames).

The user can also select any phrase and be shown similar info about it in the phrase info panel. That is, they are not limited to the phrases picked out as candidates by the software.

==

Clicking on a highlighted or selected phrase (in the item) does two things:

• side effect on the model: marks it as "being a topic phrase in the item" (in some display modes, applied to this item in the future, this makes that phrase permanently highlighted; it also has an effect on item classification, associating that phrase-instance with the "informal word-based topic" corresponding to its words.)

• temporary side effect: selects it (as text in the item).

(While there is any text selection, the phrase info panel is locked to displaying info about that phrase, ignoring mouseover.)

While a phrase is selected, clicking on one of its disambiguations shown in the phrase info panel marks it as not only being a topic phrase (whether or not it was already so marked), but as referring to that specific topic.

(The topic phrases exist in a tree (really a DAG) of more specific variants. The formal ones (sort of like urls) are the most specific; informal word-based ones are least specific; words with disambiguation info (like the names of wikipedia pages of the form "A (B)") are in between.)

==

If the user's desired more-specific topic phrase is not shown in the phrase info panel, the user can get it there (finding it or making a new one) in various ways, by doing things in that panel:

• search for it among known topic phrases (using a search field)

• see more candidates than are shown by default (using a scroll bar or by clicking "show more" or so)

• make up a new topic phrase, e.g. by typing one

• paste a topic phrase (as text, with a special syntax if it's formal) which they got from somewhere else

Once it's there, they can click on it (to associate it with the selected phrase in the item) as if it had been there all along; or for some ways of making it get there, that action would already be done implicitly.

==

In general, about actions in the phrase info panel:

Actions in the panel don't affect the item content.

Most of them don't affect the text selection in the item, but some might, e.g. after finding a new topic there might be a minibutton or context-menu-item on it to "select its occurrences in the item".

Some of them affect whether certain item text is marked as being a topic phrase, and if so, which formal topic it's marked as referring to.

==

The semantics of item text that is marked as referring to a specific topic is similar to the semantics of item text marked as being in a certain font, in a rich text UI, but not exactly the same.

Editing unmarked text doesn't change which text is marked, or how.

Editing the marked text itself would not change its font in a rich text UI, but that would probably be an unexpected behavior if it's marked as meaning a topic, since it might no longer make sense.

So what it does instead is leave it marked as "having a former meaning as a certain topic". (Exception: if the edited phrase in the item was already known to sometimes have the same meaning as the original, it would not make this change.)

Some user (not necessarily the same one who edits the phrase in the item) can then disambiguate that in two ways -- make it a real topic mark again (in various ways), or just delete this marking.

==

deleting a topic-marking

In the topic-classifying UI, for any marked topic phrase, MAYBE there is a minibutton near it (maybe it looks like a red [x]) which can be used to delete its topic-marking.

clicking on the phrase itself (as implied above) brings its info into the phrase info panel so you can edit the topic marking. Probably there is also a way in that panel to delete it. MAYBE that will be sufficient (for deleting these markings) and we won't need any minibutton near the marked topic text in the item after all.

===

The UI abilities outlined above are already enough to put items into a graph based on relatedness, since two items are indirectly connected whenever they share topics, and topics can be as fine-grained and/or formal as anyone likes.

==

TODO: add more UI info and stories about more direct inter-item links, and/or directional or typed links to topics.

Add stories which show how an item can be classified as an argument in favor or against a certain relation between topics, e.g. "thorium reactors" are or are not relevant to "reducing human-produced CO2", and "reducing human-produced CO2" is or is not "highly desirable". (Where it's more useful if the topics are more formal, but the stories make sense either way; we also want stories about how to make topics more formal.)

==