📌
Classification is one of the important applications in different fields, which is used in each field according to the need. Categorizing questions is also used in different departments, for example, by categorizing questions, you can prioritize their answers, or, for example, by categorizing them based on gender, you can choose who will answer the question, etc. The task of categorizing questions can be done automatically. Here we have created several classifiers with the help of different methods and using deep networks, which can be used depending on different applications.
1. Classification by gender:
This classification was taught on the questions on an Arabic site where the questioner s gender was known, but since many questions are not conceptually dependent on gender, it cannot be expected to be very accurate. But according to the testing of different methods and estimation of the output, in the best case, we reached an accuracy of about 66%.
2. Classification based on age:
This category was taught on Arabic questions in which the age of the questioner was known. At first, the age of people was divided into 10-year intervals (10 age intervals from 1 to 100 years), and by testing different methods, we reached an accuracy of about 35%. In the second stage, we divided people into 4 life periods (adolescence period up to 19 years old, youth period from 20 to 32 years old, middle age period from 33 to 50 years old and old age period from 51 years old and above). Using this data, various methods were tested, and considering that many questions do not depend on the age of people, the output was obtained at about 45%.
3. Classification based on tags in Arabic:
At first, it was necessary to create a suitable dataset with limited tags to create a classification. For this purpose, we first came to 5 tags (Al-Salat, Al-Khams, Al-Qur an and Al-Hadith, Al-Soom and Al-Tahart) from several thousand different tags, and we packed many existing tags in these 5 tags. In the next step, with the help of the existing questions that were packaged in these 5 categories, different classifications were made with the help of different methods, and in the best method, we reached about 92% accuracy.
4. Classification based on tags in Farsi language:
At first, it was necessary to create a suitable dataset with limited tags to create a classification. For this purpose, we first found 8 tags from several thousand different tags (morality and mysticism, political and social, Quran and hadith, history and tradition, beliefs and theology, Mahdism, rulings and theology), which included many of the existing tags in this We packed 8 tags. In the next step, with the help of the existing questions that were packaged in these 8 categories, different classifications were made with the help of different methods, and in the best method, we reached about 93% accuracy.
