语意情感分析和Gephi图分析

第一部分

在这个博客中，实现了一个简单的语意情感分析，将文件中的所有单词，逐句进行分析，并对每个句子给出一个评分。1表示正面，0表示中立，-1表示负面。

其中用到了语意字典，拿到某个名次或动词后，要在字典中查找其感情色彩，并将该值记录。每个句子输出一个分数。

代码如下：

import nltk
from nltk.tokenize import sent_tokenize
import csv
import numpy

def getposneg(searchword):
    wordcsv_file = open("wordcsv_simp.csv","r")

    pos = []
    neg = []
    found = 0
    for row in csv.DictReader(wordcsv_file):
        if row['Word'] == searchword:
            pos.append(float(row['Pos']))
            neg.append(float(row['Neg']))
            found = 1

    if found == 0:
        return 0
    else:
        #print "pos=%s, neg=%s"%(numpy.mean(pos),numpy.mean(neg))
        if numpy.mean(pos) - numpy.mean(neg) > 0.1:
            return 1
        elif numpy.mean(pos) - numpy.mean(neg) < -0.1:
            return -1
        else:
            return 0

    wordcsv_file.close()

file = open('1155099711.txt')
text = file.read()
filename = open('sentiment_scores.txt','a')

text = text.replace(",",".")

# tokenize opinion into sentences. Try to break the sentences by comma as well
sent_tokenize_list = sent_tokenize(text)
lenSent = len(sent_tokenize_list)#remember the counts of sentences

#negative_prefix_list,for checking the pre words of adj so to find if it is not positive
negative_prefix_list = ['no','not','nor','none','never','nothing','neither','hardly','barely','rarely','scarcely','little','few','seldom']

# examine each sentence one by one
for k in range(lenSent):
    total_score = 0
    #print sent_tokenize_list[k]
    filename.write(sent_tokenize_list[k]+'\n')
    text_tok = nltk.word_tokenize(sent_tokenize_list[k])

    Tag = nltk.pos_tag(text_tok)
    #print "Number of tag = %s"%len(Tag)
    lenTag = len(Tag)

    np = [];
    flag = [1 for i in range(lenTag)]#set all the flag of Tag is 1
    nlist = [];

    # try to group several nouns 'NN' / 'NNP' into noun phrases
    for t in range(lenTag):
        if flag[t] == 1:
            if Tag[t][1] == "NN":
                if Tag[t+1][1] == "NN":
                    np.append(text_tok[t]+" "+ text_tok[t+1])
                    flag[t+1] = 0
                    flag[t] = 0
                    nlist.append(t+1)
                else:
                    flag[t] = 0
                    np.append(text_tok[t])
                    nlist.append(t)
            elif Tag[t][1] == "NNP":
                if Tag[t+1][1] == "NNP" and Tag[t+2][1] == "NN":
                    flag[t] = 0
                    flag[t+1] = 0
                    flag[t+2] = 0
                    np.append(text_tok[t]+" "+text_tok[t+1]+" "+ text_tok[t+2])
                    nlist.append(t+2)
                elif Tag[t+1][1] == "NN":
                    flag[t] = 0
                    flag[t+1] = 0
                    np.append(text_tok[t]+" "+text_tok[t+1])
                    nlist.append(t+1)
                else:
                    flag[t] = 0
                    np.append(text_tok[t])
                    nlist.append(t)

    #print "List of noun=%s"%np
    #print "Corresponding number=%s"%nlist

    # Find sentiment + check negation
    jlist = [];
    sentlist = [];#sentiments of adj and verbs
    for t in range(lenTag):
        if Tag[t][1] == "JJ" or Tag[t][1] =='JJS' or Tag[t][1] == 'JJR':   # find adjectives
            jlist.append(t)
            #print 'adj',text_tok[t]
            sentiment = getposneg(text_tok[t])
            #print 'adj sentiment',sentiment
            if text_tok[t-1] in negative_prefix_list or text_tok[t-2] in negative_prefix_list: # you should add more negative words
                sentiment *= -1
            total_score = total_score + sentiment # add every single sentiment score to total score
            if sentiment > 0:
                jj_sentiment = "positive"
            elif sentiment < 0:
                jj_sentiment = "negative"
            else:
                jj_sentiment = "neutral"
            sentlist.append(jj_sentiment)
            #print "Adj: %s at %s is %s"%(Tag[t][0],t,jj_sentiment)
    if len(jlist) == 0:  # if there is no adjectives, find verbs that may have sentiments
        for t in range(lenTag):
            if (Tag[t][1] == "VB" or  Tag[t][1] == "VBD" or Tag[t][1] == "VBG" or Tag[t][1] == "VBN" or Tag[t][1] == "VBP" or Tag[t][1] == "VBZ"):   
                #print 'verb word',text_tok[t]
                sentiment = getposneg(text_tok[t])
                total_score = total_score + 0.5*sentiment
                #print 'verb sentiment',sentiment
                if sentiment > 0:
                    verb_sentiment = "positive"
                elif sentiment < 0:
                    verb_sentiment = "negative"
                else:
                    verb_sentiment = "neutral"
                sentlist.append(verb_sentiment)
                #print "Verb: %s at %s is %s"%(Tag[t][0],t,verb_sentiment)

    if total_score > 0:
        total_score = 1
    elif total_score <0:
        total_score = -1
    else:
        total_score = 0

    #print 'Total score of this sentence: ',total_score
    input_string = 'Total score of this sentence: ' + str(total_score) +'\n' +"---------------------------------------------"+'\n\n'

    #with open(filename,'a') as f:
    filename.write(input_string)

file.close()

第二部分

这部分使用node之间的关系，将40个人组成的小社团之间的社交关系进行了分析，生成node文件和edge文件，并导入到Gephi中，绘制出他们之间的社交关系网络图。

代码如下：

import networkx as nx
import csv

h = nx.DiGraph()
for i in range(1,41):
    h.add_node(i)

blog = list(csv.reader(open('blogpost4.csv','r')))

for i in range(1,len(blog)):
    line = blog[i]
    for j in range(len(line)):
        if line[j] !='0':
            h.add_edge(i,j+1,weight=int(line[j]))

with open('edges.csv','w') as csvfile:
    writer = csv.writer(csvfile)
    writer.writerow(['Row','Column','Weight'])
    for (u,v,d) in h.edges(data='weight'):
        writer.writerow([u,v,d])


with open('N.csv','w') as csvfile:
    writer = csv.writer(csvfile)
    writer.writerow(['Id','Label'])
    for node in h.nodes():
        writer.writerow([int(node),'U'+str(node)])

with open('E.csv','w') as csvfile:
    writer = csv.writer(csvfile)
    writer.writerow(['Source','Target','Type','Weight'])
    for (u,v,d) in h.edges(data='weight'):
        writer.writerow([u,v,'Directed',d])


print 'density:',nx.density(h)

avg_indegree = 0
for key,value in nx.in_degree_centrality(h).items():
    avg_indegree = avg_indegree + value
    if key == 33:
        print 'My own in-degree:',value
avg_indegree = float(avg_indegree/40)
print 'avg_indegree:',avg_indegree

#print 'outdegree centrality',nx.out_degree_centrality(h)
avg_outdegree = 0
for key,value in nx.out_degree_centrality(h).items():
    avg_outdegree = avg_outdegree + value
    if key == 33:
        print 'My own out-degree:',value
avg_outdegree = float(avg_outdegree/40)
print 'avg_outdegree:',avg_outdegree

#print 'betweenness centrality',nx.betweenness_centrality(h)
avg_betweenness = 0
for key,value in nx.betweenness_centrality(h).items():
    avg_betweenness = avg_betweenness + value
    if key == 33:
        print 'My own betweenness centrality:',value
avg_betweenness = avg_betweenness/40
print 'avg_betweenness:',avg_betweenness

#print 'closeness centrality:',nx.closeness_centrality(h)
avg_closeness = 0
for key,value in nx.closeness_centrality(h).items():
    avg_closeness = avg_closeness + value
    if key == 33:
        print 'My own closeness centrality',value
avg_closeness = avg_closeness/40
print 'avg_closeness:',avg_closeness

社交关系图如下：