AI Zone Admin Forum Add your forum

NEWS: Chatbots.org survey on 3000 US and UK consumers shows it is time for chatbot integration in customer service!read more..

Identify Personality Type within a simple conversation
 
 

So, I have a working theory that you can identify a Myers-Briggs Personality Type, based on a simple short conversation. The more you talk, the more confident you can become on the personality type…..

Some background….
Myers-Briggs Personality Type.  Wiki MB Definition
For example, an ISTJ is someone who fits the categories of Introvert, Sensor, Thinker and Judger. This personality type, ISTJ, will think before they speak.  You’ll notice that they quietly reflect before responding.  Their conversation is very sequential and you can usually follow their thought process easily because they are concerned with making sure everything is as linear and factual as possible.  They will use concrete details.  You will see a lot of comparative words in their speech, such as “looks like”, “is similar to”, or “remember when”.  They can easily recall past data and experiences and will pull from that to add authority to what they’re saying.

The theory being, if you know the basic personality of who you are talking to,  you would tailor your conversation engagement style to suit the person.

For example,
Typically, an ISTJ values logic, rules and consistency.
ISTJs are more grounded in logic and facts than in emotions. Because of this, ISTJs tend to be less aware of their own feelings and less considerate or aware of the feelings of others.
Don’t take it personally if an ISTJ seems cold or doesn’t pick up on cues for reassurance or emotional understanding.  If you need to talk about feelings with an ISTJ, spell them out.
Because ISTJs are more at home with practical realities than they are with the abstract and theoretical, they will be more comfortable with areas of conversation that reflect this tendency. You can expect to see an ISTJ relating anecdotes and narrative stories from her life rather than discussing ideas in broader, more unspecific terms. If this is different from your own tendencies, look for the ways in which stories can relate ideas, both yours and hers.

In ChatBot land, this seems like it could be a reality to identify the person you are talking to and to respond in a manner that is smart and effective.

So, to test this theory. (Identify the personality type) I took some very big datasets of conversations that are pre-classified as personality types. Each dataset is a bunch of conversations purported to be for people of a specific personality type.

And I threw it into a program which identified the word count incidence of each word.
After a few hours, I was able to generate a very large array of word, personality type and incidence rate. Here is a subset of this, for the word “I”:


TYPE Word %
ESFJ I 0.058682
INFP I 0.058360
ENFP I 0.057930
ESTJ I 0.054533
ESFP I 0.054471
INFJ I 0.054230
INTP I 0.053500
ESTP I 0.051630
INTJ I 0.051520
ENTJ I 0.047900
ENTP I 0.047840
ENFJ I 0.044270

IN this sample, ESFJ people say the word “I” the most. ENFJ say it the least. With more samples, this may change.  But this was done for ALL the words in each sample.

How to easily identify the personality type:

It would seem to me that it would be simple to just read in the incoming words.
For each word, look up the % based on the personality type, and add the % to a personality array. This array would be the total % for each personality type.

Add up all the %, and the type with the highest number is likely the personality type.
With more data, it would be more and more accurate.

With the personality type in hand, this would be used to tailor a chatbot response that would be better received by the recipient.

just some random thoughts…. PM me if you want to collab on this.

 

 
  [ # 1 ]

I do find this interesting but I can’t spare the time to work on it, so I’ll just add some thoughts.

There has been research on this sort of thing in psychology. For example, it is surmised that you can estimate someone’s intelligence/education from their length of sentences and frequent use of link words. Women can supposedly also be distinguished through more frequent use of pronouns.
https://www.scientificamerican.com/article/you-are-what-you-say

This could certainly be helpful for chatbots. I have for instance somewhere in my program a choice of responding to a problem statement either through problem-solving or with a sympathetic response. A classic example of conversational misalignment is when a man gives advice to a woman who only wants to feel heard. (Instead of “woman”, it would be more accurate to say “a social person”). If one can establish that the user has a “social” personality, then changing a global setting in a chatbot could give greater weight to sympathetic responses over advice, and thus avoid such misalignment. Or in simpler chatbots one could set it as a condition for certain responses.
Having said that, I think current chatbots have bigger problems than social finetuning to fix first.

Wouldn’t it be more accurate to analyse the “E/S/F/J” personality traits individually rather than the “ESFJ” personality types as a whole? Telling an introvert from an extrovert is an easier distinction to make than 12 mixes of traits.

I think you can only tell personality from common words, like stopwords, link words and reporting verbs. Most other words will be an indication of which topic is discussed rather than personality. In addition, factors like word length, rarity of words, and sentence length, could be interesting indicators. For instance people with autism, typically introvert, tend to have larger than average vocabularies.

 

 
  [ # 2 ]

Totally agree that the current chatbots have bigger problems to solve first, other than fine-tuning social nuances. On my end, this is a challenge as well.

yes, excellent point on conducting analysis on the specific “E/S/F/J” personality traits individually rather than the “ESFJ” personality types as a whole. I suppose the data could be grouped at a higher level.

Instead of grouping at the 16 lower levels, they can be grouped at a higher level.
extraversion vs introversion (8 vs 8).
sensing vs intuition (8 v 8)
thinking vs feeling
judgment vs perception

Also, I did notice that for many words, there was no statistical correlation between personality types. so, this supports your theory on using specific types of words.

As for a practical application, chatbot responses to important questions would be influenced by the personality type of the person talking to the avatar.  Within cs, i am keeping a running total to determine the dominant personality, and using this to influence the responses for key questions.
cheers

 

 
  login or register to react