File size: 2,815 Bytes
883ea44
 
9df8406
883ea44
 
9df8406
 
 
 
 
 
 
 
a1e484f
9df8406
 
 
 
a1e484f
 
 
 
 
9df8406
 
883ea44
 
9df8406
38d87ea
883ea44
9df8406
38d87ea
 
883ea44
38d87ea
 
 
 
 
 
 
 
883ea44
38d87ea
883ea44
 
 
 
38d87ea
 
 
 
 
 
 
 
 
 
 
 
 
9df8406
 
 
 
883ea44
9df8406
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
from lime.lime_text import LimeTextExplainer
from nltk.tokenize import sent_tokenize
from predictors import predict_for_explainanility


def explainer(text, model_type):
    def predictor_wrapper(text):
        return predict_for_explainanility(text=text, model_type=model_type)

    class_names = ["negative", "positive"]
    explainer_ = LimeTextExplainer(
        class_names=class_names, split_expression=sent_tokenize
    )
    sentences = [sent for sent in sent_tokenize(text)]
    num_sentences = len(sentences)
    exp = explainer_.explain_instance(
        text, predictor_wrapper, num_features=num_sentences, num_samples=500
    )
    weights_mapping = exp.as_map()[1]
    sentences_weights = {sentence: 0 for sentence in sentences}
    for idx, weight in weights_mapping:
        if 0 <= idx < len(sentences):
            sentences_weights[sentences[idx]] = weight
    print(sentences_weights, model_type)
    return sentences_weights, exp


def analyze_and_highlight(text, model_type):

    highlighted_text = ""
    sentences_weights, _ = explainer(text, model_type)
    positive_weights = [weight for weight in sentences_weights.values() if weight >= 0]
    negative_weights = [weight for weight in sentences_weights.values() if weight < 0]

    smoothing_factor = 0.001  # we do this cos to avoid all white colors
    min_positive_weight = min(positive_weights) if positive_weights else 0
    max_positive_weight = max(positive_weights) if positive_weights else 0
    min_negative_weight = min(negative_weights) if negative_weights else 0
    max_negative_weight = max(negative_weights) if negative_weights else 0

    max_positive_weight += smoothing_factor
    min_negative_weight -= smoothing_factor

    for sentence, weight in sentences_weights.items():
        sentence = sentence.strip()
        if not sentence:
            continue

        if weight >= 0 and max_positive_weight != min_positive_weight:
            normalized_weight = (weight - min_positive_weight + smoothing_factor) / (
                max_positive_weight - min_positive_weight
            )
            color = f"rgb(255, {int(255 * (1 - normalized_weight))}, {int(255 * (1 - normalized_weight))})"
        elif weight < 0 and min_negative_weight != max_negative_weight:
            normalized_weight = (weight - max_negative_weight - smoothing_factor) / (
                min_negative_weight - max_negative_weight
            )
            color = f"rgb({int(255 * (1 - normalized_weight))}, 255, {int(255 * (1 - normalized_weight))})"
        else:
            color = "rgb(255, 255, 255)"  # when no range

        highlighted_sentence = (
            f'<span style="background-color: {color}; color: black;">{sentence}</span> '
        )
        highlighted_text += highlighted_sentence

    return highlighted_text