| import jieba |
| import numpy as np |
| import json |
| from tqdm import tqdm |
| corpus_prob_dict ={} |
|
|
|
|
| def calcDocP(tokens): |
| doc_len = len(tokens) |
| freq = {} |
| for token in tokens: |
| freq[token] = freq.get(token, 0) + 1 |
| prob = {} |
| for word in freq.keys(): |
| prob[word] = freq[word] / doc_len |
| return prob |
|
|
| def calcCorpusP(corpus): |
| tokens = [] |
| for doc in corpus: |
| tokens.extend(doc["content"]) |
| corpus_len = len(tokens) |
| freq = {} |
| for token in tokens: |
| freq[token] = freq.get(token, 0) + 1 |
| prob = {} |
| for word in freq.keys(): |
| prob[word] = freq[word] / corpus_len |
| return prob |
|
|
| def querylikelihood(doc, query_tokens,lam): |
| doc_prob_dict = calcDocP(doc["content"]) |
| log_ql = 1 |
| for word in query_tokens: |
| doc_prob = doc_prob_dict.get(word, 0) |
| corpus_prob = corpus_prob_dict.get(word, 0) |
| |
| smooth_prob = lam * doc_prob + (1 - lam) * corpus_prob |
| log_prob = np.log(smooth_prob+1e-100) |
| log_ql += log_prob |
| |
| return log_ql |
|
|
| def topk_docs(docs, query, lam, k): |
| ql_list = [] |
| |
| for doc in docs: |
| ql = querylikelihood(doc, query, lam) |
| ql_list.append((doc["id"], ql)) |
| ql_list.sort(key=lambda x: x[1], reverse=True) |
| return [x[0] for x in ql_list[:k]] |
|
|
|
|
| stopwords = ['\n','\\','n',' ','.','（','）'] |
|
|
| with open('../../data/example/cn_stopwords.txt',encoding='UTF-8') as f: |
| lines=f.readlines() |
| for line in lines: |
| stopwords.append(line.strip()) |
|
|
| testids=[] |
|
|
| with open('../../data/example/dev.query.txt', "r", encoding="utf-8") as file: |
| for line in file: |
| qid,q = line.split('\t') |
| testids.append(int(qid)) |
|
|
|
|
| with open('../../data/queries.json', "r", encoding="utf-8") as file: |
| data = json.load(file) |
|
|
| docs=[] |
|
|
| with open('../../data/corpus.jsonl',encoding='UTF-8') as f: |
| for line in f: |
| doc = json.loads(line) |
| words = list(jieba.cut(doc['content'])) |
| words = [word for word in words if word not in stopwords] |
| |
| new_doc = {} |
| new_doc['id']=doc['id'] |
| new_doc['content']=words |
| docs.append(new_doc) |
|
|
| corpus_prob_dict = calcCorpusP(docs) |
|
|
| lam = 0.9 |
|
|
| tot = 0.0 |
| one_recall_hits = 0.0 |
| recall_hits = 0.0 |
| mrr_sum = 0.0 |
|
|
| topk = 3 |
|
|
| for obj in tqdm(data): |
| if(type(obj['问题'])==float): |
| continue |
| if(obj['query_id'] not in testids): |
| continue |
| tot += 1 |
| q = list(jieba.cut(obj['问题'])) |
| q = [word for word in q if word not in stopwords] |
| ans = topk_docs(docs, q, lam, topk) |
| |
| for rank,match in enumerate(ans, 1): |
| if match in obj['match_id']: |
| one_recall_hits += 1 |
| mrr_sum += 1.0/rank |
| break |
| cnt = 0 |
| for match in obj['match_id']: |
| if match in ans: |
| cnt += 1 |
| recall_hits += cnt/len(obj['match_id']) |
|
|
| print(f"topk:{topk}") |
| print(f'one_recall:{one_recall_hits/tot}') |
| print(f'Recall:{recall_hits/tot}') |
|
|
| print(f'mrr:{mrr_sum/tot}') |
| |
