Huti - Q & A Service
There is a lot of controversy about copycats (repeating the original idea, without making changes). Is it worth repeating the Western counterparts, if so, is it worth veiling and at least modifying the interface. There is no consensus on this. We decided to repeat one of the western startups.
Many probably heard about the Aardvark service , and about its subsequent purchase by Google. We did not immediately repeat it. They waited a while to see if Google would adapt it. When we realized that no steps were being taken in this direction, we decided to try ourselves in social search, especially since we regularly used Aardvark and faced with the problem of the lack of support for the Russian language and the lack of a sufficient number of users from the Russian-speaking region.
So huti.ru appeared . It is based on the same idea as Aardvark’s - a social search in which you can ask a question, and the service itself determines the topic of the question and looks for suitable people who can answer it. To get started, you need to register and indicate your expert areas. Next, you need to add a bot to your contacts, which you can ask a question of interest at any time.
The algorithm that we put in Huti is:
1. The user asks a question.
2. The system extracts keywords based on frequency characteristics similar to TFIDF .
3. Next, the system builds a list of topics with weights that correspond to the selected words and phrases. The knowledge about the topics and words that describe them was obtained on the basis of public information that includes Google answers, Yandex answers, Wikipedia, and DMOS .
4. Next, the system searches for users whose topics are most relevant to the topics and weights of the issue.
5. This weight is multiplied by the user's characteristics - how often he answers, how often his answers suit those who ask, how much their age matches, etc.
6. The system polls users one by one until someone answers and until the person who asked responds to the answer. Thus, the smallest number of people is involved in this process and (as planned) the concern of the system participants is minimal.
7. The question closes as soon as a sufficient number of satisfactory answers is received or the maximum of people who saw the question is exceeded. This is how we ensure the quick closing of irrelevant questions, as well as those questions that have already been answered.
The main problem that was encountered in the initial stages and which is still relevant is the insufficient core of users and topics. And since the system in any case finds people who, in its opinion, can answer it, there are times when the user found thematically is actually quite far from the topic of the question. We hope to solve this problem by increasing the number of users and, accordingly, expanding the thematic base.
To date, the statistics are as follows:
Users: 600
Questions: 1700
Answers: 4300
Average speed of receiving an answer to a question: 30 min.
Supported messengers: GTalk, Jabber, ICQ (not always online). There was also a bot for Facebook chat, but it was banned, apparently they suspected something was amiss.
We will be happy if the service helps someone quickly find answers to questions, or if you are ready to share your knowledge with others.
To register, you can use the site. After that, you need to indicate your Messenger contact on the corresponding page and then add the Hooti bot to your contacts (Huti will not knock on you even after he finds out your contact).
Or go a faster huti-way - add a new contact to your messenger (contact list here - huti.ru/staticpages/about_better), send him the text habr_huti_invitation, and go through the quick registration in the question-answer mode.
Update: Due to the website’s shortcomings, we recommend that users of GTalk and Jabber so far register and communicate with the Hooti using the messenger. How to log in to the site for those who registered through the messenger, see paragraph 5 - huti.ru/staticpages/about_help
Many probably heard about the Aardvark service , and about its subsequent purchase by Google. We did not immediately repeat it. They waited a while to see if Google would adapt it. When we realized that no steps were being taken in this direction, we decided to try ourselves in social search, especially since we regularly used Aardvark and faced with the problem of the lack of support for the Russian language and the lack of a sufficient number of users from the Russian-speaking region.
So huti.ru appeared . It is based on the same idea as Aardvark’s - a social search in which you can ask a question, and the service itself determines the topic of the question and looks for suitable people who can answer it. To get started, you need to register and indicate your expert areas. Next, you need to add a bot to your contacts, which you can ask a question of interest at any time.
The algorithm that we put in Huti is:
1. The user asks a question.
2. The system extracts keywords based on frequency characteristics similar to TFIDF .
3. Next, the system builds a list of topics with weights that correspond to the selected words and phrases. The knowledge about the topics and words that describe them was obtained on the basis of public information that includes Google answers, Yandex answers, Wikipedia, and DMOS .
4. Next, the system searches for users whose topics are most relevant to the topics and weights of the issue.
5. This weight is multiplied by the user's characteristics - how often he answers, how often his answers suit those who ask, how much their age matches, etc.
6. The system polls users one by one until someone answers and until the person who asked responds to the answer. Thus, the smallest number of people is involved in this process and (as planned) the concern of the system participants is minimal.
7. The question closes as soon as a sufficient number of satisfactory answers is received or the maximum of people who saw the question is exceeded. This is how we ensure the quick closing of irrelevant questions, as well as those questions that have already been answered.
The main problem that was encountered in the initial stages and which is still relevant is the insufficient core of users and topics. And since the system in any case finds people who, in its opinion, can answer it, there are times when the user found thematically is actually quite far from the topic of the question. We hope to solve this problem by increasing the number of users and, accordingly, expanding the thematic base.
To date, the statistics are as follows:
Users: 600
Questions: 1700
Answers: 4300
Average speed of receiving an answer to a question: 30 min.
Supported messengers: GTalk, Jabber, ICQ (not always online). There was also a bot for Facebook chat, but it was banned, apparently they suspected something was amiss.
We will be happy if the service helps someone quickly find answers to questions, or if you are ready to share your knowledge with others.
To register, you can use the site. After that, you need to indicate your Messenger contact on the corresponding page and then add the Hooti bot to your contacts (Huti will not knock on you even after he finds out your contact).
Or go a faster huti-way - add a new contact to your messenger (contact list here - huti.ru/staticpages/about_better), send him the text habr_huti_invitation, and go through the quick registration in the question-answer mode.
Update: Due to the website’s shortcomings, we recommend that users of GTalk and Jabber so far register and communicate with the Hooti using the messenger. How to log in to the site for those who registered through the messenger, see paragraph 5 - huti.ru/staticpages/about_help