martinthenext

docs = [
    '''About us. We deliver Artificial Intelligence & Machine Learning
       solutions to solve business challenges.''',
    '''Contact information. Email [martin davtyan at filament dot ai]
       if you have any questions''',
    '''Filament Chat. A framework for building and maintaining a scalable
       chatbot capability''',
]

martinthenext

scorer.feature_weights = [0.6, 0.4]
scorer.score(query)

array([0.18856181, 0.23136585, 0.15411592])

docs[scorer.score(query).argmax()]

'Contact information. Email [martin davtyan at filament dot ai] if you have any questions'

martinthenext

scorer.learn_feedback(feedback)
scorer.score(query)

array([0.94280904, 0.69987944, 0.25685987])

docs[scorer.score(query).argmax()]

'About us. We deliver Artificial Intelligence & Machine Learning solutions to solve business challenges.'

martinthenext

docs[scorer.score(query).argmax()]

'Filament Chat. A framework for building and maintaining a scalable chatbot capability'

martinthenext

scorer = Scorer(docs)
query

'who is making chatbots information'

scorer.score(query)

array([0.        , 0.22847492, 0.25685987])

martinthenext

class Scorer():
    """ Scores documents for a search query based on tf-idf
        similarity and relevance feedback
        
    """
    def __init__(self, docs):
        """ Initialize a scorer with a collection of documents, fit a 
            vectorizer and list feature functions
        
        """

martinthenext

nn_similarity = np.max(similarity)
pos_feedback_feature = [nn_similarity * pos_feedback_proportions.get(idx, 0.) 
                        for idx, _ in enumerate(docs)]
pos_feedback_feature

[0.4714045207910317, 0.23570226039551584, 0.0]

martinthenext

from collections import Counter

counts = Counter(pos_feedback_doc_idx)
counts

Counter({0: 2, 1: 1})

pos_feedback_proportions = {
        doc_idx: count / sum(counts.values()) for doc_idx, count in counts.items()
}

martinthenext

pos_feedback_doc_idx = [idx for idx, feedback_value 
                        in feedback[feedback_queries[max_idx]] 
                        if feedback_value == 1.]
pos_feedback_doc_idx

[0, 1, 0]

martinthenext

import numpy as np

query = 'who is making chatbots information'
feedback_queries = list(feedback.keys())

similarity = cosine_similarity(vectorizer.transform([query]), 
                               vectorizer.transform(feedback_queries))
similarity

array([[0.70710678, 0.        ]])