首页
学习
活动
专区
工具
TVP
发布
精选内容/技术社群/优惠产品,尽在小程序
立即前往
您找到你想要的搜索结果了吗?
是的
没有找到

【干货】基于注意力机制的神经匹配模型用于短文本检索

【导读】在基于检索的问答系统中,很重要的一步是将检索到的答案进行排序得到最佳的答案。在检索到的答案比较短时,对答案进行排序也成为了一个难题。使用深度学习的方法,如建立在卷积神经网络和长期短期记忆模型基础上的神经网络模型,不需要手动设计语言特征,也能自动学习问题与答案之间的语义匹配,但是缺陷是需要词汇重叠特征和BM25等附加特征才能达到较好的效果。本文分析了出现这个问题的原因,并提出了基于值的权值共享的神经网络,并使用注意力机制为问题中的值赋予不同的权值。专知内容组编辑整理。 论文: aNMM: Rankin

08

【HDU 4940】Destroy Transportation system(无源无汇带上下界可行流)

Tom is a commander, his task is destroying his enemy’s transportation system. Let’s represent his enemy’s transportation system as a simple directed graph G with n nodes and m edges. Each node is a city and each directed edge is a directed road. Each edge from node u to node v is associated with two values D and B, D is the cost to destroy/remove such edge, B is the cost to build an undirected edge between u and v. His enemy can deliver supplies from city u to city v if and only if there is a directed path from u to v. At first they can deliver supplies from any city to any other cities. So the graph is a strongly-connected graph. He will choose a non-empty proper subset of cities, let’s denote this set as S. Let’s denote the complement set of S as T. He will command his soldiers to destroy all the edges (u, v) that u belongs to set S and v belongs to set T.  To destroy an edge, he must pay the related cost D. The total cost he will pay is X. You can use this formula to calculate X:

01
领券