Using Hubs and Authorities to Make Chatting Robots More Interesting
Currently, there is a problem with chatting robots. They don’t bring up tangential conversational threads to the current conversation when the conversation becomes stale. Humans by instinct do bring up related tangents to keep the conversation going when the current topic of the conversation reaches an ending point with no other things to talk about.
For example, say a person and a robot are talking about college football. The person could be from Cornell and be talking about how their team lost to Brown in a long, grueling game that went to double overtime. Currently robots are smart enough to say, “Oh wow, that must have been a disappointing game.” Then the person could possibly say, “Yeah, sometimes I feel like it would be better for the team to just lose big when they lose and ease the pain of having to watch a close defeat. Uh, so yeah…” This is an example of a stalling conversation. Most of the time, people would probably change the topic to something else that is related to the current topic. Say, for example, how the NFL team they root for is going. However, a typical chatting robot of today is not able to do this task which is so simple and instinctual for humans to do. They do not have the ability to make the conversation more interesting through bringing up tangents.
The article below talks about how scientists are trying to improve chatting robots so that they do bring up these tangential threads. In order to do this, they have to solve two problems: the problem of making the robots know when there is a stalled conversation while making sure they don’t constantly change the subject, which would be annoying, and the problem of knowing what tangents to bring up that are related to the current conversation. The focus of this article and where the hubs and authorities algorithm comes in is in solving the second problem of knowing what tangents to bring up. There needs to be a resource where the robot can take from in order to tangent. The obvious answer to this is the Internet.
Two algorithms are used in order to extract ideas for tangents to bring up in the conversation. The first is Google’s PageRank algorithm or whatever algorithm Google uses now-a-days in their search engine in order to find the most popular and relevant web pages related to the current topic of conversation. The second is the Hubs and Authorities Algorithm, which is also known as HITS (Hyperlink-Induced Topic search). This algorithm is used in order to analyze the results returned by the first algorithm for usability in the conversation.
As seen in class, HITS runs on the subset of pages returned by Google’s PageRank, not the entire web. Unlike PageRank, HITS treats each page as either a hub or an authority, not both. The official HITS algorithm returns the authorities in ranked order of authority score. The hub scores are used simply to compute the authority scores. However, for this case, it could be useful to return the hubs in ranked order. The authorities will contain information about the current conversational topic. On the other hand, the hubs contain links to the current conversational topic. These hubs are the ones that are most likely to contain valuable information on possible tangents to the current conversational topic. This is just one of the major applications of the HITS algorithm that the PageRank algorithm alone can’t solve.
Why chatting with robots might become more interesting in the future