3 ms·
LDA/Topic Modeling is interesting stuff. I always feel like the way this data gets surfaced as "tags" is very ineffective. Any non-tech person would look at thi
by gibrown 12y ago
LDA/Topic Modeling is interesting stuff. I always feel like the way this data gets surfaced as "tags" is very ineffective. Any non-tech person would look at this and generally be confused. So this item is triggering my rants against tagging:
- Tagging is like trying to predict the future. What word will help some future person to get to this content?
- Tagging often tries to fill the hole left by bad search
- There is no evaluation method to measure how good a set of tags are
- Tags make very bad UI clutter.
Some of these points are related to encouraging users to tag content, but auto-tagging also seems problematic.
To me something more along the lines of entity extraction is more useful because it is a well defined problem, and can be used to improve a lot of other applications.
- sitkack 12y agoIt seems like you would want to run k-means over the comments and the tags to pull out semantically meaningful words for tags and then reduce the total number of tags over the corpus. Then say, use wikipedia to generate an automatic taxonomy where those extract words occur.
- doppenhe 12y agowe do have k-means in our API as well this wouldn't be hard to do.
- sytelus 12y agoTagging is useful to summarize the content. It's like saying describe this article in 3 nouns. Lot of HN articles are cryptic and if you can pull out good tags it can be helpful to prioritizing its reading. It's even more helpful when there are 100s of comments and you want to know key topics of discussion. The problem is often that generated tags are often very poor quality. To understand the utility of tagging, look at some article, read it and then put 3 words that best describes the topics. I bet most others would find human generated tags very useful. Machine generated tags are usually no where close to what humans would generate.
- lugg 12y agoI would like to see this kind of tagging being used to improve search results and simply hiding them. Could increase rank when there are more tag hits. Although I guess that is essentially what good search is. Tags while cluttering the ui do help you find similar content. Still not as good as a good reccomendation system but a decent stop gap measure in some instances.