詞語、句子相似度比較

synonyms自帶的相似度比較結果並不能滿意。
以下提高了名詞分數占比，隨便寫的，錯誤之處請指正

 

主要代碼

# encoding=utf8
import synonyms,jieba,re

jieba.load_userdict('mydict.txt')
sen1 = "硅膠成品的銷售及生產"
temp2 = ['橡膠加工專用設備制造', '石墨及碳素制品制造', '其他橡膠制品制造', '建築材料生產專用機械制造', '肉制品及副產品加工', "特種陶瓷制品制造"]

def getWordType(word):
    w,t = synonyms.seg(word)
    return t[0]

# 去重、排序
def deal_list(objs, order='name', isLen=True, reverse=False):
    temp,result = {},[]
    for obj in objs:
        temp[obj[order]] = obj
    for prpo in temp:
        result.append(temp[prpo])
    if isLen:
        return sorted(result, key=lambda obj: len(obj[order]))
    return sorted(result, key=lambda obj: obj[order], reverse=reverse)

def wordCompare(instr, sentences):
    result = []
    keys = {}
    words,types =  synonyms.seg(instr)
    for i,w1 in enumerate(words):
        keys[w1] = []
        if re.findall('[vn]+', types[i]) and len(w1)>1:
            ws,ss = synonyms.nearby(w1)
            for j,w2 in enumerate(ws):
                if ss[j]<0.6:
                    break
                keys[w1].append({
                    'text': w2,
                    'source': ss[j],
                    'typeSource': 2 if re.findall('[vn]+', getWordType(w2)) else 0.5
                })
        if len(keys[w1])==0:
            keys[w1].append({
                'text': w1,
                'source': 1,
                'typeSource': 2 if re.findall('[vn]+', getWordType(w1)) else 0.5 #名詞類得分加倍，其它對折
            })

    for i in sentences:
        source = 0
        for j in keys:
            bfSource = 1 #降低同一個詞的同義詞影響
            for k in keys[j]:
                if k['text'] in i:
                    source += (bfSource*k['source']*k['typeSource'])
                    bfSource = k['source']
        result.append({
            'text': i,
            'source': source
        })
    return result

if __name__ == '__main__':
    temp2 = deal_list(wordCompare(sen1, temp2), order='source', isLen=False, reverse=True)
    for i in temp2:
        print sen1, i['text'], i['source']

 

 

mydict.txt

酒 1000000 n
膠 10000000 n

 

運行結果