Integrated entity and relation extractions with BERT to target_extractor with...

Integrated entity and relation extractions with BERT to target_extractor with good results on unseen guitar reviews.

ADA/server/agent/target_extraction/BERT/entity_extractor/bert_entity_extractor.py

@@ -7,6 +7,7 @@ from torch.optim import Adam
 import time
 import numpy as np
 from sklearn import metrics
+import statistics
 from transformers import get_linear_schedule_with_warmup
 from agent.target_extraction.BERT.entity_extractor.entity_dataset import EntityDataset, generate_batch, generate_production_batch
 from agent.target_extraction.BERT.entity_extractor.entitybertnet import NUM_CLASSES, EntityBertNet
@@ -175,8 +176,34 @@ class BertEntityExtractor:
         recall = metrics.recall_score(targets, outputs, average=None)
         print('recall:', recall)
 
+    def extract_entity_probabilities(self, terms, file_path=None, dataset=None, size=None):
+        # load data
+        if file_path is not None:
+            data, _ = EntityDataset.from_file(file_path, size=size)
+        else:
+            if dataset is None:
+                raise AttributeError('file_path and data cannot both be None')
+            data = dataset
+
+        loader = DataLoader(data, batch_size=BATCH_SIZE, shuffle=False, num_workers=4,
+                            collate_fn=generate_production_batch)
+
+        self.net.cuda()
+        self.net.eval()
+
+        probs = {term: [] for term in terms}
+
+        with torch.no_grad():
+            for input_ids, attn_mask, entity_indices, instances in loader:
+                # send batch to gpu
+                input_ids, attn_mask, entity_indices = tuple(i.to(device) for i in [input_ids, attn_mask,
+                                                                                    entity_indices])
+
+                # forward pass
+                output_scores = softmax(self.net(input_ids, attn_mask, entity_indices), dim=1)
+                entity_scores = output_scores.narrow(1, 1, 1).flatten()
+
+                for ins, score in zip(instances, entity_scores.tolist()):
+                    probs[ins.entity].append(score)
 
-extr: BertEntityExtractor = BertEntityExtractor.train_and_validate('camera_backpack_laptop_review_entities.tsv',
-                                                             'trained_bert_entity_extractor_camera_backpack_laptop.pt',
-                                                             valid_frac=0.05,
-                                                             valid_file_path='annotated_acoustic_guitar_review_entities.tsv')
\ No newline at end of file
+        return {t: statistics.mean(t_probs) if len(t_probs) > 0 else None for t, t_probs in probs.items()}

ADA/server/agent/target_extraction/BERT/entity_extractor/entity_dataset.py

@@ -9,7 +9,7 @@ from agent.target_extraction.BERT.relation_extractor.pairbertnet import TRAINED_
 
 MAX_SEQ_LEN = 128
 LABELS = ['ASPECT', 'NAN']
-LABEL_MAP = {'ASPECT': 1, 'NAN': 0}
+LABEL_MAP = {'ASPECT': 1, 'NAN': 0, None: None}
 MASK_TOKEN = '[MASK]'
 tokenizer = BertTokenizer.from_pretrained(TRAINED_WEIGHTS)
 
@@ -34,7 +34,7 @@ def generate_production_batch(batch):
     input_ids = encoded['input_ids']
     attn_mask = encoded['attention_mask']
 
-    entity_indices = indices_for_entity_ranges([instance.range for instance in batch])
+    entity_indices = indices_for_entity_ranges([instance.entity_range for instance in batch])
 
     return input_ids, attn_mask, entity_indices, batch
 

ADA/server/agent/target_extraction/BERT/relation_extractor/bert_rel_extractor.py

@@ -196,7 +196,7 @@ class BertRelExtractor:
 
             print(instances[0].get_relation_for_label(output_labels[0]))
 
-    def extract_relations(self, file_path=None, dataset=None, size=None):
+    def extract_relations(self, n_aspects, aspect_index_map, aspect_counts, file_path=None, dataset=None, size=None):
         # load data
         if file_path is not None:
             data, _ = PairRelDataset.from_file(file_path, size=size)
@@ -211,7 +211,8 @@ class BertRelExtractor:
         self.net.cuda()
         self.net.eval()
 
-        outputs = []
+        prob_matrix = np.zeros((n_aspects, n_aspects))
+        count_matrix = np.zeros((n_aspects, n_aspects))
 
         with torch.no_grad():
             for input_ids, attn_mask, masked_indices, prod_indices, feat_indices, instances in loader:
@@ -223,24 +224,18 @@ class BertRelExtractor:
 
                 # forward pass
                 output_scores = softmax(self.net(input_ids, attn_mask, masked_indices, prod_indices, feat_indices), dim=1)
-                _, output_labels = torch.max(output_scores.data, 1)
-
-                outputs += map(lambda x: x[0].get_relation_for_label(x[1]), zip(instances, output_labels.tolist()))
-
-                for ins, scores, out in zip(instances, output_scores.tolist(), output_labels.tolist()):
-                    print(ins.text)
-                    print(ins.tokens)
-                    print(scores)
-                    print(ins.get_relation_for_label(out))
-                    print('---')
+                rel_scores = output_scores.narrow(1, 1, 2)
 
-        return outputs
+                for ins, scores in zip(instances, rel_scores.tolist()):
+                    forward_score, backward_score = scores
+                    fst_idx, snd_idx = aspect_index_map[ins.fst], aspect_index_map[ins.snd]
+                    prob_matrix[snd_idx][fst_idx] += forward_score
+                    prob_matrix[fst_idx][snd_idx] += backward_score
+                    count_matrix[snd_idx][fst_idx] += 1
+                    count_matrix[fst_idx][snd_idx] += 1
 
+        prob_matrix = (prob_matrix.T / aspect_counts).T  # scale rows by aspect counts
 
-extr: BertRelExtractor = BertRelExtractor.train_and_validate('../data/camera_backpack_laptop_review_pairs_no_nan.tsv',
-                                                             'trained_bert_rel_extractor_camera_backpack_laptop_no_nan.pt',
-                                                             size=10000,
-                                                             valid_frac=0.05,
-                                                             valid_file_path='data/annotated_acoustic_guitar_review_pairs.tsv')
+        return prob_matrix
 
 

ADA/server/agent/target_extraction/target_extractor.py

@@ -2,49 +2,50 @@ import pandas as pd
 from collections import Counter
 from nltk import pos_tag
 from nltk.tokenize import word_tokenize, sent_tokenize
-from nltk.corpus import stopwords, wordnet
+from nltk.corpus import stopwords
 from nltk.stem import WordNetLemmatizer
 import string
 from gensim.models.phrases import Phrases, Phraser
-from anytree import Node
+from anytree import Node, RenderTree
 import numpy as np
 import re
 from gensim.models import Word2Vec
 import pickle
 from agent.target_extraction.product import Product
+from agent.target_extraction.BERT.entity_extractor.entity_dataset import EntityDataset
+from agent.target_extraction.BERT.entity_extractor.bert_entity_extractor import BertEntityExtractor
 from agent.target_extraction.BERT.relation_extractor.pair_rel_dataset import PairRelDataset
 from agent.target_extraction.BERT.relation_extractor.bert_rel_extractor import BertRelExtractor
 from pathos.multiprocessing import ProcessingPool as Pool
 import itertools
 
-np.set_printoptions(precision=4, threshold=np.inf, suppress=False)
+np.set_printoptions(precision=3, threshold=np.inf, suppress=True)
 stop_words = stopwords.words('english')
 wnl = WordNetLemmatizer()
 sentiment_lexicon = pd.read_csv('data/NRC-Sentiment-Lexicon-Wordlevel-v0.92.tsv', sep='\t', index_col=0)
-bert_extractor_path = 'BERT/trained_bert_rel_extractor_camera_and_backpack_with_nan.pt'
-pool = Pool(4)
+entity_extractor_path = 'BERT/entity_extractor/trained_bert_entity_extractor_camera_backpack_laptop.pt'
+rel_extractor_path = 'BERT/relation_extractor/trained_bert_rel_extractor_camera_backpack_laptop_no_nan.pt'
+pool = Pool(2)
 
 
 class TargetExtractor:
-    PHRASE_THRESHOLD = 4
-    MIN_RELATEDNESS = 0.3
     N_ASPECTS = 100
-    MIN_DIRECT_GAIN = 0.1
-    DEPTH_COST = 0.3
-    FREQ_OVER_PARENT = 10  # target must appear x times more frequently than in parent
-    OUTLIER_COEFFICIENT = 5
-    N_DIRECT_FEATURES = 3  # top N_DIRECT_FEATURES features will be direct children of the product (not subfeatures)
-    PARENT_COUNT_FRAC = 0.5  # feature f1 will only be considered as a subfeature of f2 if c(f1) / c(f2) > this value
-    WV_SIZE = 100
-    WV_WINDOW = 9
+
+    # phraser
+    PHRASE_THRESHOLD = 4
+
+    # tree
+    SUBFEATURE_MULT = 2  # for z to be a subfeature of x, matrix(z, x) > matrix(z, f) * SUBFEATURE_MULT for all other f
 
     # word2vec
-    MIN_TERM_COUNT = 0
-    MIN_SIMILARITY = 0
-    SYNONYM_SIMILARITY = 0.1
+    MIN_TERM_COUNT = 100
+    SYNONYM_SIMILARITY = 0.10
+    WV_SIZE = 100
+    WV_WINDOW = 7
 
     # bert
     MAX_BERT_DATASET_SIZE = 300000
+    ENTITY_PROB_THRESHOLD = 0.5
 
     # parent is a TargetExtrator of a parent category, eg. > electronics > camera
     def __init__(self, product, file_path, text_column):
@@ -53,7 +54,7 @@ class TargetExtractor:
 
         print('tokenizing phrases...')
         # tokenize and normalize phrases
-        texts = TargetExtractor.obtain_texts(file_path, text_column, n=50000)
+        texts = TargetExtractor.obtain_texts(file_path, text_column)
         self.sentences = list(itertools.chain.from_iterable(pool.map(sent_tokenize, texts)))
         self.sentences = pool.map(lambda s: s.replace('_', ' ').lower(), self.sentences)
         self.phrases = pool.map(word_tokenize, self.sentences)
@@ -81,7 +82,7 @@ class TargetExtractor:
 
         print('mining aspects...')
         # mine aspects
-        aspects, counts = self.get_related_nouns(self.counter)
+        aspects, counts = self.get_aspects(self.counter)
 
         print('extracting synonyms...')
         # obtain synonyms
@@ -97,21 +98,22 @@ class TargetExtractor:
         self.save()
 
         print('extracting relatedness matrix...')
-        # extract relationships between aspects
         self.relatedness_matrix = self.get_bert_relations()
 
-        print(self.aspects)
-        print(self.syn_dict)
-        print(self.relatedness_matrix)
+        self.save()
 
-        # print('extracting aspect tree...')
-        # # extract aspect tree
-        # self.tree = self.get_product_tree()
+        print('extracting aspect tree...')
+        self.tree = self.get_product_tree()
 
-        # print('saving...')
+        print('saving...')
         self.save()
-        #
-        # print('done:')
+
+        print('done:')
+
+        print(self.aspects)
+        print(self.syn_dict)
+        print(self.relatedness_matrix)
+        print(RenderTree(self.tree))
 
     def extract_relatedness_matrix(self):
         print('extracting relatedness matrix...')
@@ -155,7 +157,7 @@ class TargetExtractor:
             if len(term) > 1 and (any(not re.compile('NN|JJ').match(tag) for _, tag in term)
                                   or any(tag.startswith('JJ') and polar_adjective(t) for t, tag in term)):
                 return [subterm for subterm, _ in term]
-            return ['_'.join([subterm for subterm, _ in term])]
+            return [' '.join([subterm for subterm, _ in term])]
 
         def polar_adjective(adj):
             return adj in sentiment_lexicon.index and (sentiment_lexicon.loc[adj]['positive'] == 1 or
@@ -164,48 +166,21 @@ class TargetExtractor:
         return [subterm for term in tagged_unfiltered for subterm in filter_ngram(term)]
 
     def get_bert_relations(self):
-        print('  select phrases for BERT...')
+        print('  select phrases for relation extraction...')
         pair_texts = [rel for rel in pool.map(self.pair_relations_for_text, range(len(self.phrases))) if rel is not None]
-        bert_df = pd.DataFrame(pair_texts, columns=['sentText', 'relationMentions'])
+        df = pd.DataFrame(pair_texts, columns=['sentText', 'relationMentions'])
 
         print('  extracting relations with BERT...')
-        dataset = PairRelDataset.from_df(bert_df, size=TargetExtractor.MAX_BERT_DATASET_SIZE)
-        bert_extractor = BertRelExtractor.load_saved(bert_extractor_path)
-        rels = bert_extractor.extract_relations(dataset=dataset)
-
-        print('  calculating relatedness matrix...')
-        positive_matrix = np.zeros((len(self.aspects), len(self.aspects)))
-        negative_matrix = np.zeros((len(self.aspects), len(self.aspects)))
-
-        for e1, rel, e2 in rels:
-            idx_super = self.index_for_aspect(e1)
-            idx_sub = self.index_for_aspect(e2)
-
-            if idx_super is None or idx_sub is None:
-                raise AttributeError('Could not find an aspect for term', e1 if idx_super is None else e2)
-            if rel not in ['/has_feature', '/no_relation']:
-                raise AttributeError('Could not recognize relation', rel)
-
-            if rel == '/has_feature':
-                positive_matrix[idx_sub][idx_super] += 1
-            else:
-                negative_matrix[idx_sub][idx_super] += 1
-
-        relatedness_matrix = (positive_matrix - negative_matrix) / (positive_matrix + negative_matrix)
-
-        # scale values by counts of aspects
-        # counts_arr = np.zeros(len(self.aspects))
-        # for idx, aspect in enumerate(self.aspects):
-        #     counts_arr[idx] = self.counts[aspect]
-        # relatedness_matrix = (relatedness_matrix.T / counts_arr).T
-
-        # scale linearly to be more readable
-        # relatedness_matrix = relatedness_matrix / relatedness_matrix.max()
+        dataset = PairRelDataset.from_df(df, size=TargetExtractor.MAX_BERT_DATASET_SIZE)
+        bert_extractor = BertRelExtractor.load_saved(rel_extractor_path)
+        aspect_counts = np.array([self.counts[aspect] for aspect in self.aspects])
+        relatedness_matrix = bert_extractor.extract_relations(len(self.aspects), self.aspect_index_map(), aspect_counts,
+                                                              dataset=dataset)
 
         return relatedness_matrix
 
-    def index_for_aspect(self, aspect):
-        return next(idx for idx, a in enumerate(self.aspects) if aspect.lower().replace(' ', '_') in self.syn_dict[a])
+    def aspect_index_map(self):
+        return {syn: idx for idx, aspect in enumerate(self.aspects) for syn in self.syn_dict[aspect]}
 
     def pair_relations_for_text(self, idx):
         text = self.sentences[idx]
@@ -218,15 +193,14 @@ class TargetExtractor:
 
         found_aspects = []
         for aspect in self.aspects:
-            found_aspect = False
+            found_form = False
             for form in self.syn_dict[aspect]:
                 if form in ngrams:
-                    form_with_space = form.replace('_', ' ')
-                    if len(found_aspects) > 1 or found_aspect or overlapping_terms(found_aspects, form_with_space):
+                    if len(found_aspects) > 1 or found_form or overlapping_terms(found_aspects, form):
                         # cannot have more than two aspects, or two forms of the same aspect, or overlapping terms
                         return None
-                    found_aspects.append(form_with_space)
-                    found_aspect = True
+                    found_aspects.append(form)
+                    found_form = True
 
         return (text, [{'em1Text': found_aspects[0], 'em2Text': found_aspects[1]}]) if len(found_aspects) == 2 else None
 
@@ -251,8 +225,8 @@ class TargetExtractor:
         nouns = []
         word_idx = 0
         for token in ngrams:
-            if '_' in token:
-                words = token.split('_')
+            if ' ' in token:
+                words = token.split(' ')
                 word_range = range(word_idx, word_idx + len(words))
                 has_noun = any(is_noun(pos_tags[i]) for i in word_range)
                 all_terms_valid = all(is_valid_term(pos_tags[i]) for i in word_range)
@@ -267,30 +241,61 @@ class TargetExtractor:
                 word_idx += 1
         return nouns
 
-    def get_related_nouns(self, counter):
+    def get_aspects(self, counter):
+        # take N_ASPECTS most common terms
         term_counts = counter.most_common()[:TargetExtractor.N_ASPECTS]
-
         terms = [term for term, count in term_counts]
-        # bring product to front of list
-        if self.product in terms:
-            terms.remove(self.product)
-        else:
-            terms.pop()
-        terms.insert(0, self.product)
 
-        return terms, {term: count for term, count in term_counts}
+        print('  preparing entity texts for BERT...')
+        entity_texts = [t for t in pool.map(lambda i: self.entity_mentions_in_text(i, terms),
+                                            range(len(self.sentences)))
+                        if t is not None]
+        df = pd.DataFrame(entity_texts, columns=['sentText', 'entityMentions'])
+
+        print('  extracting entities with BERT...')
+        dataset = EntityDataset.from_df(df, size=TargetExtractor.MAX_BERT_DATASET_SIZE)
+        entity_extractor = BertEntityExtractor.load_saved(entity_extractor_path)
+        probs = entity_extractor.extract_entity_probabilities(terms, dataset=dataset)
+
+        aspects = [term for term in terms if probs[term] is not None and probs[term] >= TargetExtractor.ENTITY_PROB_THRESHOLD]
 
-    def is_related_to_product(self, term, wv):
-        return wv.similarity(self.product, term) > TargetExtractor.MIN_SIMILARITY
+        # bring product to front of list
+        if self.product in aspects:
+            aspects.remove(self.product)
+        aspects.insert(0, self.product)
+
+        return aspects, {term: count for term, count in term_counts if term in aspects}
+
+    def entity_mentions_in_text(self, text_idx, entities):
+        text = self.sentences[text_idx]
+        all_tokens = set().union(*[self.phrases[text_idx], self.ngram_phrases[text_idx]])
+
+        entity_mention = None
+        for entity in entities:
+            n_mentions = sum(1 for token in all_tokens if entity == token.lower())
+            if n_mentions > 1:
+                # many mentions of same entity
+                return None
+            if n_mentions == 1:
+                if entity_mention is None:
+                    entity_mention = entity
+                elif entity_mention in entity:
+                    entity_mention = entity
+                elif entity not in entity_mention:
+                    # text cannot have more than one entity mention, unless one is a subset of the other,
+                    # in which case the longer one is taken
+                    return None
+
+        if entity_mention is not None:
+            return text, [{'text': entity_mention}]
+
+        return None
 
     @staticmethod
     def get_syn_pairs(terms, model):
         return {frozenset((t1, t2)) for t1 in terms for t2 in terms
                 if t1 != t2 and model.relative_cosine_similarity(t1, t2) > TargetExtractor.SYNONYM_SIMILARITY}
 
-    def frequency_for_term(self, term):
-        return self.counter[term] / self.total_count
-
     def get_word2vec_model(self, size, window, min_count):
         model = Word2Vec(self.ngrams(self.phrases), size=size, window=window, min_count=min_count).wv
         return model
@@ -307,83 +312,24 @@ class TargetExtractor:
         f.close()
         return extractor
 
-    @staticmethod
-    def wordnet_relatedness(t1, t2):
-        fst = wordnet.synset(t1 + '.n.01')
-        snd = wordnet.synset(t2 + '.n.01')
-        return fst.wup_similarity(snd)
-
-    @staticmethod
-    def spanning_tree_from_root(vertices, weights, root_idx=0):
-        root = Node(vertices[root_idx])
-        for idx in np.flip(np.argsort(weights[root_idx])):
-            if idx == root_idx:
-                continue
-            gain = max(TargetExtractor.MIN_DIRECT_GAIN, weights[root_idx][idx])
-            parent = root
-            for branch_node in root.descendants:
-                min_scaled_weight = min(weights[n.idx][idx] * pow(TargetExtractor.DEPTH_COST, branch_node.depth)
-                                        for n in (branch_node,) + branch_node.anchestors if n != root)
-                if min_scaled_weight > gain:
-                    gain = min_scaled_weight
-                    parent = branch_node
-            node = Node(vertices[idx], parent=parent)
-            node.idx = idx
-
-        return root
-
     def get_product_tree(self):
         root = Node(self.aspects[0])
         root.idx = 0
-        for idx in range(1, TargetExtractor.N_DIRECT_FEATURES + 1):
-            node = Node(self.aspects[idx], parent=root)
-            node.idx = idx
-        unassigned = {idx for idx in range(TargetExtractor.N_DIRECT_FEATURES + 1, len(self.aspects))}
 
         for idx in range(1, len(self.aspects)):  # for each feature in order from highest to lowest count
-            print(self.aspects[idx])
-            # create node for aspect with parent root if it is unassigned
-            node = next((n for n in root.descendants if n.idx == idx), None)
-            if not node:
-                node = Node(self.aspects[idx], parent=root)
-                node.idx = idx
-                unassigned.remove(idx)
-
-            # get highest dependant from unassigned aspects if there exists a significant one
-            dep_idx = self.get_dependant(idx, [], unassigned)
-            while dep_idx is not None:
-                print('   ', self.aspects[dep_idx])
-                # assign dep as subfeature of t
-                dep_node = Node(self.aspects[dep_idx], parent=node)
-                dep_node.idx = dep_idx
-                unassigned.remove(dep_idx)
-                # get highest dependant from remaining targets if there exists a significant one
-                dep_idx = self.get_dependant(idx, [child.idx for child in node.children], unassigned)
+            row = self.relatedness_matrix[idx]
+            max_idx1, max_idx2 = row[1:].argsort()[-2:][::-1] + 1
+            if max_idx1 < idx and row[max_idx1] >= row[max_idx2] * TargetExtractor.SUBFEATURE_MULT:
+                parent = next(n for n in root.descendants if n.idx == max_idx1)
+            else:
+                parent = root
+            node = Node(self.aspects[idx], parent=parent)
+            node.idx = idx
 
         self.node_map = {n.idx: n for n in (root,) + root.descendants}
 
         return root
 
-    @staticmethod
-    def kruskal(vertices, edges):
-        result = set()
-        groups = {vertex: i for i, vertex in enumerate(vertices)}
-        for u, v in sorted(edges, key=edges.get, reverse=True):
-            if groups[u] != groups[v]:
-                result.add((u, v))
-                TargetExtractor.join_groups(groups, groups[u], groups[v])
-        return result
-
-    @staticmethod
-    def join_groups(groups, i, j):
-        for v in groups:
-            if groups[v] == j:
-                groups[v] = i
-
-    @staticmethod
-    def singular(word):
-        return wnl.lemmatize(word)
-
     @staticmethod
     def print_relations(target_indices, dep_matrix, targets):
         idx_pairs = {frozenset((idx1, idx2)) for idx1 in target_indices for idx2 in target_indices if idx1 != idx2}
@@ -400,57 +346,6 @@ class TargetExtractor:
         for rel_idx in sorted(range(len(self.aspects)), key=lambda i: rels[i], reverse=True):
             print('    {:.4f}'.format(rels[rel_idx]), self.aspects[rel_idx])
 
-    def get_dependant(self, idx, child_indices, unassigned_indices):
-        ignore_idx = [0] + child_indices
-        max_dependant = (None, 0)
-        for u_idx in unassigned_indices:
-            # print('   ', self.aspects[u_idx])
-            dependence = self.get_significant_dependence(u_idx, ignore_idx=ignore_idx)
-            if dependence is not None:
-                dep_idx, score = dependence
-                # print('           ', self.aspects[dep_idx], score)
-                if dep_idx == idx and score > max_dependant[1]:
-                    max_dependant = (u_idx, score)
-        return max_dependant[0]
-
-    def get_significant_dependence(self, idx, ignore_idx=None):
-        if not ignore_idx:
-            ignore_idx = [0]
-        deps = self.relatedness_matrix[idx].copy()
-        for i in ignore_idx:
-            deps[i] = np.ma.masked
-        fst_high_outlier = TargetExtractor.high_outlier_idx(deps)
-        # print('       ', fst_high_outlier)
-        if fst_high_outlier is not None:
-            deps[fst_high_outlier[0]] = np.ma.masked
-            snd_high_outlier = TargetExtractor.high_outlier_idx(deps)  # np.delete(deps, fst_high_outlier)
-            # print('       ', snd_high_outlier)
-            if snd_high_outlier is None:
-                return fst_high_outlier
-        return None
-
-    @staticmethod
-    def high_outlier_idx(arr):
-        q1 = np.nanquantile(np.ma.filled(arr, np.NaN), 0.25)
-        q3 = np.nanquantile(np.ma.filled(arr, np.NaN), 0.75)
-        max_idx = np.nanargmax(np.ma.filled(arr, np.NaN))
-        lim = q3 + TargetExtractor.OUTLIER_COEFFICIENT * (q3 - q1)
-        # print('       ', arr, arr[max_idx], q1, q3, lim)
-        return (max_idx, arr[max_idx] / lim) if arr[max_idx] > lim else None
-
-    @staticmethod
-    def are_correlated(idx1, idx2, dep_matrix, ignore_idx=None):
-        if not ignore_idx:
-            ignore_idx = []
-        mask = [i in ignore_idx for i in range(len(dep_matrix[idx1]))]
-        deps1 = np.ma.masked_array(dep_matrix[idx1], mask=mask)
-        deps2 = np.ma.masked_array(dep_matrix[idx2], mask=mask)
-        return TargetExtractor.in_q3(deps1, idx2) and TargetExtractor.in_q3(deps2, idx1)
-
-    @staticmethod
-    def in_q3(arr, idx):
-        return arr[idx] >= np.quantile(arr, 0.75)
-
 
 class Synset:
 
@@ -509,7 +404,4 @@ class Synset:
         return None
 
 
-extr: TargetExtractor = TargetExtractor('backpack', 'data/verified_backpack_reviews.tsv', 'reviewText')
-# extr.print_relations_from('keys')
-# print('')
-# extr.print_relations_from('keyboard')
+extr: TargetExtractor = TargetExtractor('guitar', 'data/verified_acoustic_guitar_reviews.tsv', 'reviewText')