Saurabh askedon {{::getFormatedLocalTime("2018-10-22T12:29:11.677Z", {without_time: true})}}
Any openings in data science and machine learning? LinkedIn: https://www.linkedin.com/in/saurabh-kamble-52791314a/
I have 12 months of Experience with Python[Numpy,Pandas,MatplotlibSeaborn,Sklearn] and R programming. I have worked on Machine Learning , Natural Language Processing, Deep learning, Data Visualization tools like Tableau , QlikSense , Advance Excel and MySQL, looking for applying these skills on real-time data.Technical Skills:Applied Statistical Modelling for models analysing Z-score, Skew and Kurtosis, Data Transformation, Hypothesis Testing, Using ML Algorithms like Linear/Logistic Regression, Decision Trees, Random Forest,Xgboost,Naive-Bayes,PCA,SVM,K-means,KNN including working on Natural Language Processing with TensorFlow. Sentiment Analysis on IMDB movie dataset, Image Classification using CNN, Customer Segmentation/Churn Analysis.
Firstly you provide a user some mechanism to enter his input. That can be voice based or text based. There are various options available for voice recognition and synthesis. I used https://github.com/TalAter/annyang for speech synthesis on web and I must say, it is highly intelligent.
Then you send this transcript or user's input text to a AI/ML model which can extract meaningful keywords along with the intent from this input. You need to train the model with some sample user commands/intent. This model will then parse the text and provide you a structured form of the input. I used https://api.ai which also gives you options to integrate with various platforms.
The structured form of the input received from the AI model can be used to query the data from any data store of your liking, e.g:- mongodb, elastic search etc. You can then display the results to the user as per requirement.
Thats all you need to build a bot outta nowhere.
1. Easiest is to use tools like - IBM Watson, api.ai and wit.ai (pretty stringent, does not learn on its own, time to build and complexity is less)
2. If you wish to make a more dynamic bot and have tons on sample conversations, you can use tensorflow with a seq2seq model. (provided the right dataset and training, it will be better in conversation as compared to anything in option 1, but will be dificult for you to resolve the intent and attach data to solve the user query)
3. Hybrid - use nlp and intent detection from any service in point 1, and have custom implementation for the rest.
4. Entire custom stack. (very difficult likes of Google, Facebook, Amazon have taken years and they are not near perfection yet)
I'de suggest you start with option 1 and move down the order.