The following project contains a web application that leverages Artificial Intelligence such as Machine Learning, Natural Language Processing, and Information Retrieval. The following contains a detailed guide on how to run my project.
- PHP version 7.3.2
- Python version 3.9
- Javascript version 1.7
- Html 5
- CSS
- Ajax
- MariaDB (MySQL) version 10.1.38
- Google Chrome version 92.0.4515.159
- ApacheFriends XAMPP version 3.2.2
- PyCharm Community Edition version 2021.1.2
- SciKit-Learn version 0.24.2
- PHP Random Name Generator
- NLTK version 3.6.2
- Pandas version 1.3.0
- Numpy version 1.21.0
- Calendar
- Re
This project was built with the flexibility of running either: via the Command-line Interface (static version for testing the two algorithms) or via the web-based Graphical User Interface (GUI).
The command line version of the project has been split into two different Python scripts (files) for the two algorithms. These files can be found in the "Python" folder. Within this folder, you will see KNN-static.py, and CosineSimilarity-static.py. These are the files that use the self-titled algorithms. Both of these files depend on the "Job Description.txt" and "SmallDataset2.csv" files. The "Job Description.txt" file contains the use-case job description described in the dissertation/report. The "SmallDataset2.csv" file contains the dataset used to build the model for both algorithms. The results described in the "Results" Section of the report was produced using these files.
In order to test the experiment Feature Reduction or N-grams Manipulation on either algorithm you must go this line:
tfidf = TfidfVectorizer(stop_words='english', ngram_range=(1,1), max_features=1000)To reduce the number of features, simple change max_features to
any of the numbers shown in the "Results" table under Feature
Reduction Subsection. In order to reset to the default number of
features, simply remove the max_features parameter from the
method.
As for changing the ngrams, you do this by changing ngram_range()
to either (1,1) for unigrams, (2,2) for bigrams, (3,3) for
trigrams, (4,4) for four-grams, or (5,5) for five-grams.
The hyperparameter tuning was conducted on the KNN algorithm, which can be found in the "KNN-static.py" file. To change the number of neighbours, simply go to line:
k=5
clf = KNeighborsClassifier(n_neighbors=k)At the above line, you can change the k to the desired number of neighbours shown in the Results table.
For the Top-N recommendations experiment, this can be tested by going to the following line of the "CosineSimilarity-static.py" file:
n = 10 # Change me to change the Top-N recommendations.
results = results[1:n+1]As shown above, this is set to a default of 10. Feel free to change this number to test the results shown in the table of the Results Section.
Disclaimer: I will assume you are using a Windows Machine :)
In order to run the web-based GUI, you will need to install ApacheFriends' XAMPP. Afterwards, go to "http://localhost/phpmyadmin/", click "import" at the top of the screen. Within that window, look for the "File to import:" window pane and browse to the sql file in this repository. Sometimes, when importing an .sql file requires adding another line of code to create the database but http://www.stackoverflow.com is very helpful. Hopefully, this won't be necessary and you will be able to smoothly import the database. Nevertheless, if you have successfully loaded the database, the name "mscproject" should appear as the database name.
Next, you'll need to clone the entire repository on your local machine, into the "htdocs" folder, which is a subfolder within the XAMPP folder that is automatically created when you install ApacheFriends XAMPP.
After you have cloned everything, go to http://localhost/Content-based-Recommendation-System/signin.php in your Google Chrome browser.
You should be presented with a screen that looks like this:
Use these credentials to gain access:
Username: admin
Password: P@ssw0rd
You be directed to the dashboard page.
On the page, you will be able to see the campaign discussed in the "Use-Case" Section inside the report. From this window you can see some candidates that have been randomly generated from the "LargeDataset.csv" file which is the original dataset described in the "Resources" section of the report.
From here you can simply click "View all Candidates".
Inside this page, click the orange "Generate" button at the top to generate some candidates randomly. You will be presented with a screen to enter the number of candidates you would like to generate.
Enter a value then click done. Simply choose which algorithm you would like to use for the recommendations by clicking either of the two blue buttons at the top.
You would then be able to see results of the recommendations. For example, if you select the KNN algorithm, you should be presented with a screen that looks like this:
Feel free to email me and I will try to help. Thank you.