Academic Chatbot

Chat bot app is being made, whilst keeping the perspective of students and academic personals. There are two views of this app ? Student View and Admin View ? the former is for the students to use the features provided by our app to query regarding timetable, real-time rescheduled classes, room vaca

Project Title

Academic Chatbot

Project Area of Specialization

Artificial Intelligence

Project Summary

Chat bot app is being made, whilst keeping the perspective of students and academic personals. There are two views of this app – Student View and Admin View – the former is for the students to use the features provided by our app to query regarding timetable, real-time rescheduled classes, room vacant status, realtime announcements from university SLATE portal, policies of university, information of different courses, teacher related information. Student if in any case feels that the bot didn’t give him/her correct answer, student can send his query to admin using the feedback button present at the screen.The latter is for the admin, who will update the answer of the question in case of failure of an answer by the bot.
Features include:

- Academics related queries including University policies, course information, teacher related information and GPA policies.

- Realtime announcements from university SLATE portal
- Timetable related queries including real-time rescheduled classes and room vacant   status

- Sending query to admin panel in case of failure of answer.

- Dataset updated after admin answers the query.
- Emailing academics office in case of failure of answer.
- Point of contact for parents.

Project Objectives

1) To address Academics related queries including University policies, course information, teacher related information and GPA policies.

2) Realtime announcements from university SLATE portal
3) Timetable related queries including real-time rescheduled classes and room vacant status

4)Point of contact for parents.

5) To save time of academic officers

6) To save time of students and give them information within seconds

Project Implementation Method

Data Collection

We scrapped data from university’s website i.e. nu.edu.pk using BeautifulSoup and from SLATE which is university's run-time announcements portal using Selenium. We stored this scrapped data in text files.

Moreover, we also extracted data from timetable, teacher and course allocation files by writing generic code. These files were in excel format which we collected from SLATE and supervisor respectively.

Data Preprocessing

1) Data scrapped from university’ website was pre-processed by removing special characters from it. Moreover, data from tables in university website is scrapped and then converted into meaningful English sentences.

2) Data scrapped from SLATE was pre-processed by removing special characters from it. Moreover, appending announcements upload time is also appended with the scrapped announcement.

3) Data extracted from excel sheets of timetable and course allocation files was also pre-processed by removing raw characters / special characters from it. Moreover, data extracted from the cells of excel sheets was then converted into meaningful English sentences. For example, “AI is being held at 10:30 to 12:00 in third floor room 302 on Monday”.???????

Model

We are using Gensim Doc2Vec unsupervised model; where single sentence from excel files, complete announcement from SLATE and paragraph from nu.edu.pk is being given as a single document respectively to the Doc2Vec.

Doc2Vec model takes a set of documents as an input, and then form vector embeddings of those documents that is mapped onto multiple dimensions. The user question is also converted into a vector embedding from the same trained model. Then top documents having same cosine similarity are fetched by using cosine similarity function of Doc2Vec.

We manually tune the hyper-parameters to get the best possible / accurate outcome. Once tuned and trained, our model is ready to answer the questions.???????

Handling irrelevant responses

If user finds that the response is irrelevant to his/her query or is wrong, the user can notify the chatbot by using ‘Didn’t find the answer?’ feature. In this case, the original query is added to the database which in our case is Firebase and immediately the admin is being notified about the query.

The user also receives the query id that has been assigned by the system for further inquiry about the query status.

Admin’s response to the query is directly received by the server and it immediately stores that answer in the dataset which ensures the auto-feedback approach.

Benefits of the Project

1) It will save time of academic officers and students

2) It will reduce the physical interaction between students and academic staff

3) University information and run-time information will be easily accessed and within no time

Technical Details of Final Deliverable

Data Collection

We scrapped data from university’s website i.e. nu.edu.pk using BeautifulSoup and from SLATE which is university's run-time announcements portal using Selenium. We stored this scrapped data in text files.

Moreover, we also extracted data from timetable, teacher and course allocation files by writing generic code. These files were in excel format which we collected from SLATE and supervisor respectively.

Data Preprocessing

1) Data scrapped from university’ website was pre-processed by removing special characters from it. Moreover, data from tables in university website is scrapped and then converted into meaningful English sentences.

2) Data scrapped from SLATE was pre-processed by removing special characters from it. Moreover, appending announcements upload time is also appended with the scrapped announcement.

3) Data extracted from excel sheets of timetable and course allocation files was also pre-processed by removing raw characters / special characters from it. Moreover, data extracted from the cells of excel sheets was then converted into meaningful English sentences. For example, “AI is being held at 10:30 to 12:00 in third floor room 302 on Monday”.???????

Model

We are using Gensim Doc2Vec unsupervised model; where single sentence from excel files, complete announcement from SLATE and paragraph from nu.edu.pk is being given as a single document respectively to the Doc2Vec.

Doc2Vec model takes a set of documents as an input, and then form vector embeddings of those documents that is mapped onto multiple dimensions. The user question is also converted into a vector embedding from the same trained model. Then top documents having same cosine similarity are fetched by using cosine similarity function of Doc2Vec.

We manually tune the hyper-parameters to get the best possible / accurate outcome. Once tuned and trained, our model is ready to answer the questions.???????

Handling irrelevant responses

If user finds that the response is irrelevant to his/her query or is wrong, the user can notify the chatbot by using ‘Didn’t find the answer?’ feature. In this case, the original query is added to the database which in our case is Firebase and immediately the admin is being notified about the query.

The user also receives the query id that has been assigned by the system for further inquiry about the query status.

Admin’s response to the query is directly received by the server and it immediately stores that answer in the dataset which ensures the auto-feedback approach.

Final Deliverable of the Project

Software System

Type of Industry

Education , IT

Technologies

Artificial Intelligence(AI)

Sustainable Development Goals

Quality Education, Industry, Innovation and Infrastructure

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com