根据书单来制作item2vec

数据和工具

1. glove：http://nlp.stanford.edu/projects/glove/

2. 书单数据：待审核

代码

# -*- coding: utf-8 -*-

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from __future__ import unicode_literals
import numpy as np
from scipy import stats
import sys
import os
import math
import json
import heapq


#VecFile = './models/fasttext_model_0804_09_cbow.vec'
VecFile = '../glove/vectors.txt'


class minHeap():
    def __init__(self, k):
        self._k = k
        self._heap = []

    def add(self, item):
        if len(self._heap) < self._k:
            self._heap.append(item)
            heapq.heapify(self._heap)
        else:
            if item > self._heap[0]:
                self._heap[0] = item
                heapq.heapify(self._heap)

    def get_min(self):
        if len(self._heap) > 0:
            return self._heap[0]
        else:
            return -2

    def get_all(self):
        return self._heap

def similarity(v1, v2):
    n1 = np.linalg.norm(v1)
    n2 = np.linalg.norm(v2)
    return np.dot(v1, v2) / n1 / n2

def load_vectors(input_file=VecFile):
    vectors = {}
    with open(VecFile) as fopen:
        fopen.readlines()
        for line in fopen:
            line_list = line.strip().split()
            movie_name = line_list[0]
            vec = np.array([float(_) for _ in line_list[1:]], dtype=float)
            if not movie_name in vectors:
                vectors[movie_name] = vec
    return vectors

def topk_like(cur_movie_name, k=5, print_log=False):
    global vectors
    min_heap = minHeap(k)
    like_candidates = []
    #logger.debug('vecotrs size=%d' % (len(vectors)))
    #logger.debug('cur_movie_name %s, %s' % (cur_movie_name, type(cur_movie_name)))
    if cur_movie_name not in vectors.keys():
        print(cur_movie_name+" not in vector")
        return []
    cur_vec = vectors[cur_movie_name]
    if print_log:
        print('%s top %d likes:' % (cur_movie_name, k))
    for movie_name, vec in vectors.items():
        if movie_name == cur_movie_name:
            continue
        sim = similarity(cur_vec, vec)
        if len(like_candidates) < k or sim > min_heap.get_min():
            min_heap.add(sim)
            like_candidates.append((movie_name, sim))
    if print_log:
        for t in sorted(like_candidates, reverse=True, key=lambda _:_[1])[:k]:
            print('\t%s %f' % (t[0],t[1]))
    return sorted(like_candidates, reverse=True, key=lambda _:_[1])[:k]

vectors = load_vectors(VecFile)

if __name__ == '__main__':
    movie_names = [u'追风筝的人', u'看见', u'长尾理论', u'从0到1']
    for movie_name in movie_names:
        topk_like(movie_name, print_log=True)
    #generate_movie_topk_like_result('./output/leancloud_movie_fasttext.json', k=25)

效果

追风筝的人 top 5 likes:
	HTML 0.433426
	陆小凤传奇 0.423955
	百年孤独 0.408164
	战胜华尔街 0.403829
	沉默的大多数 0.390571
看见 top 5 likes:
	雪山飞狐 0.418104
	鹿鼎记（全五册） 0.407155
	概念力 0.405503
	梦里花落知多少 0.377425
	股票大作手操盘术 0.353401
长尾理论 top 5 likes:
	怪诞行为学 0.664615
	全球32位顶尖广告文案的写作之道 0.460249
	人人都是产品经理 0.429592
	傲慢与偏见 0.412715
	七堂极简物理课 0.357686
从0到1 top 5 likes:
	创业维艰 0.696322
	浪潮之巅 0.477051
	诛仙 0.358522
	Scikit-Learn 0.354489
	Book 0.334162