v3 / modules /text_analysis /semantic_analysis-16-10-2024.py
AIdeaText's picture
Rename modules/text_analysis/semantic_analysis.py to modules/text_analysis/semantic_analysis-16-10-2024.py
86d6da4 verified
raw
history blame
16.4 kB
# modules/text_analysis/semantic_analysis.py
# [Mantener todas las importaciones y constantes existentes...]
import streamlit as st
import spacy
import networkx as nx
import matplotlib.pyplot as plt
import io
import base64
from collections import Counter, defaultdict
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity
import logging
logger = logging.getLogger(__name__)
# Define colors for grammatical categories
POS_COLORS = {
'ADJ': '#FFA07A', 'ADP': '#98FB98', 'ADV': '#87CEFA', 'AUX': '#DDA0DD',
'CCONJ': '#F0E68C', 'DET': '#FFB6C1', 'INTJ': '#FF6347', 'NOUN': '#90EE90',
'NUM': '#FAFAD2', 'PART': '#D3D3D3', 'PRON': '#FFA500', 'PROPN': '#20B2AA',
'SCONJ': '#DEB887', 'SYM': '#7B68EE', 'VERB': '#FF69B4', 'X': '#A9A9A9',
}
POS_TRANSLATIONS = {
'es': {
'ADJ': 'Adjetivo', 'ADP': 'Preposici贸n', 'ADV': 'Adverbio', 'AUX': 'Auxiliar',
'CCONJ': 'Conjunci贸n Coordinante', 'DET': 'Determinante', 'INTJ': 'Interjecci贸n',
'NOUN': 'Sustantivo', 'NUM': 'N煤mero', 'PART': 'Part铆cula', 'PRON': 'Pronombre',
'PROPN': 'Nombre Propio', 'SCONJ': 'Conjunci贸n Subordinante', 'SYM': 'S铆mbolo',
'VERB': 'Verbo', 'X': 'Otro',
},
'en': {
'ADJ': 'Adjective', 'ADP': 'Preposition', 'ADV': 'Adverb', 'AUX': 'Auxiliary',
'CCONJ': 'Coordinating Conjunction', 'DET': 'Determiner', 'INTJ': 'Interjection',
'NOUN': 'Noun', 'NUM': 'Number', 'PART': 'Particle', 'PRON': 'Pronoun',
'PROPN': 'Proper Noun', 'SCONJ': 'Subordinating Conjunction', 'SYM': 'Symbol',
'VERB': 'Verb', 'X': 'Other',
},
'fr': {
'ADJ': 'Adjectif', 'ADP': 'Pr茅position', 'ADV': 'Adverbe', 'AUX': 'Auxiliaire',
'CCONJ': 'Conjonction de Coordination', 'DET': 'D茅terminant', 'INTJ': 'Interjection',
'NOUN': 'Nom', 'NUM': 'Nombre', 'PART': 'Particule', 'PRON': 'Pronom',
'PROPN': 'Nom Propre', 'SCONJ': 'Conjonction de Subordination', 'SYM': 'Symbole',
'VERB': 'Verbe', 'X': 'Autre',
}
}
ENTITY_LABELS = {
'es': {
"Personas": "lightblue",
"Lugares": "lightcoral",
"Inventos": "lightgreen",
"Fechas": "lightyellow",
"Conceptos": "lightpink"
},
'en': {
"People": "lightblue",
"Places": "lightcoral",
"Inventions": "lightgreen",
"Dates": "lightyellow",
"Concepts": "lightpink"
},
'fr': {
"Personnes": "lightblue",
"Lieux": "lightcoral",
"Inventions": "lightgreen",
"Dates": "lightyellow",
"Concepts": "lightpink"
}
}
CUSTOM_STOPWORDS = {
'es': {
# Art铆culos
'el', 'la', 'los', 'las', 'un', 'una', 'unos', 'unas',
# Preposiciones comunes
'a', 'ante', 'bajo', 'con', 'contra', 'de', 'desde', 'en',
'entre', 'hacia', 'hasta', 'para', 'por', 'seg煤n', 'sin',
'sobre', 'tras', 'durante', 'mediante',
# Conjunciones
'y', 'e', 'ni', 'o', 'u', 'pero', 'sino', 'porque',
# Pronombres
'yo', 't煤', '茅l', 'ella', 'nosotros', 'vosotros', 'ellos',
'ellas', 'este', 'esta', 'ese', 'esa', 'aquel', 'aquella',
# Verbos auxiliares comunes
'ser', 'estar', 'haber', 'tener',
# Palabras comunes en textos acad茅micos
'adem谩s', 'tambi茅n', 'asimismo', 'sin embargo', 'no obstante',
'por lo tanto', 'entonces', 'as铆', 'luego', 'pues',
# N煤meros escritos
'uno', 'dos', 'tres', 'primer', 'primera', 'segundo', 'segunda',
# Otras palabras comunes
'cada', 'todo', 'toda', 'todos', 'todas', 'otro', 'otra',
'donde', 'cuando', 'como', 'que', 'cual', 'quien',
'cuyo', 'cuya', 'hay', 'solo', 'ver', 'si', 'no',
# S铆mbolos y caracteres especiales
'#', '@', '/', '*', '+', '-', '=', '$', '%'
},
'en': {
# Articles
'the', 'a', 'an',
# Common prepositions
'in', 'on', 'at', 'by', 'for', 'with', 'about', 'against',
'between', 'into', 'through', 'during', 'before', 'after',
'above', 'below', 'to', 'from', 'up', 'down', 'of',
# Conjunctions
'and', 'or', 'but', 'nor', 'so', 'for', 'yet',
# Pronouns
'i', 'you', 'he', 'she', 'it', 'we', 'they', 'this',
'that', 'these', 'those', 'my', 'your', 'his', 'her',
# Auxiliary verbs
'be', 'am', 'is', 'are', 'was', 'were', 'been', 'have',
'has', 'had', 'do', 'does', 'did',
# Common academic words
'therefore', 'however', 'thus', 'hence', 'moreover',
'furthermore', 'nevertheless',
# Numbers written
'one', 'two', 'three', 'first', 'second', 'third',
# Other common words
'where', 'when', 'how', 'what', 'which', 'who',
'whom', 'whose', 'there', 'here', 'just', 'only',
# Symbols and special characters
'#', '@', '/', '*', '+', '-', '=', '$', '%'
},
'fr': {
# Articles
'le', 'la', 'les', 'un', 'une', 'des',
# Prepositions
'脿', 'de', 'dans', 'sur', 'en', 'par', 'pour', 'avec',
'sans', 'sous', 'entre', 'derri猫re', 'chez', 'avant',
# Conjunctions
'et', 'ou', 'mais', 'donc', 'car', 'ni', 'or',
# Pronouns
'je', 'tu', 'il', 'elle', 'nous', 'vous', 'ils',
'elles', 'ce', 'cette', 'ces', 'celui', 'celle',
# Auxiliary verbs
'锚tre', 'avoir', 'faire',
# Academic words
'donc', 'cependant', 'n茅anmoins', 'ainsi', 'toutefois',
'pourtant', 'alors',
# Numbers
'un', 'deux', 'trois', 'premier', 'premi猫re', 'second',
# Other common words
'o霉', 'quand', 'comment', 'que', 'qui', 'quoi',
'quel', 'quelle', 'plus', 'moins',
# Symbols
'#', '@', '/', '*', '+', '-', '=', '$', '%'
}
}
##############################################################################################################
def get_stopwords(lang_code):
"""
Obtiene el conjunto de stopwords para un idioma espec铆fico.
Combina las stopwords de spaCy con las personalizadas.
"""
try:
nlp = spacy.load(f'{lang_code}_core_news_sm')
spacy_stopwords = nlp.Defaults.stop_words
custom_stopwords = CUSTOM_STOPWORDS.get(lang_code, set())
return spacy_stopwords.union(custom_stopwords)
except:
return CUSTOM_STOPWORDS.get(lang_code, set())
def perform_semantic_analysis(text, nlp, lang_code):
"""
Realiza el an谩lisis sem谩ntico completo del texto.
Args:
text: Texto a analizar
nlp: Modelo de spaCy
lang_code: C贸digo del idioma
Returns:
dict: Resultados del an谩lisis
"""
logger.info(f"Starting semantic analysis for language: {lang_code}")
try:
doc = nlp(text)
key_concepts = identify_key_concepts(doc)
concept_graph = create_concept_graph(doc, key_concepts)
concept_graph_fig = visualize_concept_graph(concept_graph, lang_code)
entities = extract_entities(doc, lang_code)
entity_graph = create_entity_graph(entities)
entity_graph_fig = visualize_entity_graph(entity_graph, lang_code)
# Convertir figuras a bytes
concept_graph_bytes = fig_to_bytes(concept_graph_fig)
entity_graph_bytes = fig_to_bytes(entity_graph_fig)
logger.info("Semantic analysis completed successfully")
return {
'key_concepts': key_concepts,
'concept_graph': concept_graph_bytes,
'entities': entities,
'entity_graph': entity_graph_bytes
}
except Exception as e:
logger.error(f"Error in perform_semantic_analysis: {str(e)}")
raise
def fig_to_bytes(fig):
buf = io.BytesIO()
fig.savefig(buf, format='png')
buf.seek(0)
return buf.getvalue()
def fig_to_html(fig):
buf = io.BytesIO()
fig.savefig(buf, format='png')
buf.seek(0)
img_str = base64.b64encode(buf.getvalue()).decode()
return f'<img src="data:image/png;base64,{img_str}" />'
def identify_key_concepts(doc, min_freq=2, min_length=3):
"""
Identifica conceptos clave en el texto.
Args:
doc: Documento procesado por spaCy
min_freq: Frecuencia m铆nima para considerar un concepto
min_length: Longitud m铆nima de palabra para considerar
Returns:
list: Lista de tuplas (concepto, frecuencia)
"""
try:
# Obtener stopwords para el idioma
stopwords = get_stopwords(doc.lang_)
# Contar frecuencias de palabras
word_freq = Counter()
for token in doc:
if (token.lemma_.lower() not in stopwords and
len(token.lemma_) >= min_length and
token.is_alpha and
not token.is_punct and
not token.like_num):
word_freq[token.lemma_.lower()] += 1
# Filtrar por frecuencia m铆nima
concepts = [(word, freq) for word, freq in word_freq.items()
if freq >= min_freq]
# Ordenar por frecuencia
concepts.sort(key=lambda x: x[1], reverse=True)
return concepts[:10] # Retornar los 10 conceptos m谩s frecuentes
except Exception as e:
logger.error(f"Error en identify_key_concepts: {str(e)}")
return [] # Retornar lista vac铆a en caso de error
def create_concept_graph(doc, key_concepts):
"""
Crea un grafo de relaciones entre conceptos.
Args:
doc: Documento procesado por spaCy
key_concepts: Lista de tuplas (concepto, frecuencia)
Returns:
nx.Graph: Grafo de conceptos
"""
try:
G = nx.Graph()
# Crear un conjunto de conceptos clave para b煤squeda r谩pida
concept_words = {concept[0].lower() for concept in key_concepts}
# A帽adir nodos al grafo
for concept, freq in key_concepts:
G.add_node(concept.lower(), weight=freq)
# Analizar cada oraci贸n
for sent in doc.sents:
# Obtener conceptos en la oraci贸n actual
current_concepts = []
for token in sent:
if token.lemma_.lower() in concept_words:
current_concepts.append(token.lemma_.lower())
# Crear conexiones entre conceptos en la misma oraci贸n
for i, concept1 in enumerate(current_concepts):
for concept2 in current_concepts[i+1:]:
if concept1 != concept2:
# Si ya existe la arista, incrementar el peso
if G.has_edge(concept1, concept2):
G[concept1][concept2]['weight'] += 1
# Si no existe, crear nueva arista con peso 1
else:
G.add_edge(concept1, concept2, weight=1)
return G
except Exception as e:
logger.error(f"Error en create_concept_graph: {str(e)}")
# Retornar un grafo vac铆o en caso de error
return nx.Graph()
def visualize_concept_graph(G, lang_code):
"""
Visualiza el grafo de conceptos.
Args:
G: Grafo de networkx
lang_code: C贸digo del idioma
Returns:
matplotlib.figure.Figure: Figura con el grafo visualizado
"""
try:
plt.figure(figsize=(12, 8))
# Calcular el layout del grafo
pos = nx.spring_layout(G)
# Obtener pesos de nodos y aristas
node_weights = [G.nodes[node].get('weight', 1) * 500 for node in G.nodes()]
edge_weights = [G[u][v].get('weight', 1) for u, v in G.edges()]
# Dibujar el grafo
nx.draw_networkx_nodes(G, pos,
node_size=node_weights,
node_color='lightblue',
alpha=0.6)
nx.draw_networkx_edges(G, pos,
width=edge_weights,
alpha=0.5,
edge_color='gray')
nx.draw_networkx_labels(G, pos,
font_size=10,
font_weight='bold')
plt.title("Red de conceptos relacionados")
plt.axis('off')
return plt.gcf()
except Exception as e:
logger.error(f"Error en visualize_concept_graph: {str(e)}")
# Retornar una figura vac铆a en caso de error
return plt.figure()
def create_entity_graph(entities):
G = nx.Graph()
for entity_type, entity_list in entities.items():
for entity in entity_list:
G.add_node(entity, type=entity_type)
for i, entity1 in enumerate(entity_list):
for entity2 in entity_list[i+1:]:
G.add_edge(entity1, entity2)
return G
def visualize_entity_graph(G, lang_code):
fig, ax = plt.subplots(figsize=(12, 8))
pos = nx.spring_layout(G)
for entity_type, color in ENTITY_LABELS[lang_code].items():
node_list = [node for node, data in G.nodes(data=True) if data['type'] == entity_type]
nx.draw_networkx_nodes(G, pos, nodelist=node_list, node_color=color, node_size=500, alpha=0.8, ax=ax)
nx.draw_networkx_edges(G, pos, width=1, alpha=0.5, ax=ax)
nx.draw_networkx_labels(G, pos, font_size=8, font_weight="bold", ax=ax)
ax.set_title(f"Relaciones entre Entidades ({lang_code})", fontsize=16)
ax.axis('off')
plt.tight_layout()
return fig
#################################################################################
def create_topic_graph(topics, doc):
G = nx.Graph()
for topic in topics:
G.add_node(topic, weight=doc.text.count(topic))
for i, topic1 in enumerate(topics):
for topic2 in topics[i+1:]:
weight = sum(1 for sent in doc.sents if topic1 in sent.text and topic2 in sent.text)
if weight > 0:
G.add_edge(topic1, topic2, weight=weight)
return G
def visualize_topic_graph(G, lang_code):
fig, ax = plt.subplots(figsize=(12, 8))
pos = nx.spring_layout(G)
node_sizes = [G.nodes[node]['weight'] * 100 for node in G.nodes()]
nx.draw_networkx_nodes(G, pos, node_size=node_sizes, node_color='lightgreen', alpha=0.8, ax=ax)
nx.draw_networkx_labels(G, pos, font_size=10, font_weight="bold", ax=ax)
edge_weights = [G[u][v]['weight'] for u, v in G.edges()]
nx.draw_networkx_edges(G, pos, width=edge_weights, alpha=0.5, ax=ax)
ax.set_title(f"Relaciones entre Temas ({lang_code})", fontsize=16)
ax.axis('off')
plt.tight_layout()
return fig
###########################################################################################
def generate_summary(doc, lang_code):
sentences = list(doc.sents)
summary = sentences[:3] # Toma las primeras 3 oraciones como resumen
return " ".join([sent.text for sent in summary])
def extract_entities(doc, lang_code):
entities = defaultdict(list)
for ent in doc.ents:
if ent.label_ in ENTITY_LABELS[lang_code]:
entities[ent.label_].append(ent.text)
return dict(entities)
def analyze_sentiment(doc, lang_code):
positive_words = sum(1 for token in doc if token.sentiment > 0)
negative_words = sum(1 for token in doc if token.sentiment < 0)
total_words = len(doc)
if positive_words > negative_words:
return "Positivo"
elif negative_words > positive_words:
return "Negativo"
else:
return "Neutral"
def extract_topics(doc, lang_code):
vectorizer = TfidfVectorizer(stop_words='english', max_features=5)
tfidf_matrix = vectorizer.fit_transform([doc.text])
feature_names = vectorizer.get_feature_names_out()
return list(feature_names)
# Aseg煤rate de que todas las funciones necesarias est茅n exportadas
__all__ = [
'perform_semantic_analysis',
'identify_key_concepts',
'create_concept_graph',
'visualize_concept_graph',
'create_entity_graph',
'visualize_entity_graph',
'generate_summary',
'extract_entities',
'analyze_sentiment',
'create_topic_graph',
'visualize_topic_graph',
'extract_topics',
'ENTITY_LABELS',
'POS_COLORS',
'POS_TRANSLATIONS'
]