人工智能培训

搜索

论文代码开源:用子词包推广词嵌入(Generalizing Word Embeddings using Bag of Subwords)

[复制链接]
admin 发表于 2018-9-15 10:10:33 | 显示全部楼层 |阅读模式
admin 2018-9-15 10:10:33 845 0 显示全部楼层
人工智能论文代码开源:用子词包推广词嵌入(Generalizing Word Embeddings using Bag of Subwords)请注意该人工智能论文代码开源在github,大部分是python写的,框架可能是tensorflow或者pytorch。当前的对话系统更多地关注文本和语音背景知识,并且通常基于两个发言者。最近的一些工作已经研究了基于静态图像的对话。然而,一些真实的人类交互也涉及动态视觉上下文(类似于视频)以及多个发言者之间的对话交换。为了更接近这种多模态对话技能和视觉应用,我们引入了一个新的视频背景,多个演讲者对话数据集,基于直播足球游戏视频和来自Twitch.tv的聊天。这个具有挑战性的测试平台允许我们开发基于视觉的对话模型,该模型应该从实时视频中生成相关的时空和空间事件语言,同时还与聊天历史相关。对于强基线,我们还提出了例如基于三向注意力流动(TriDAF)的多模式和生成模型。我们通过检索排名 - 召回,自动短语匹配指标以及人类评估研究来评估这些模型。我们还提出了数据分析,模型消融和可视化,以了解不同模态和模型组件的贡献。
Current dialogue systems focus more on textual and speech context knowledgeand are usually based on two speakers.Some recent work has investigated staticimage-based dialogue.However, several real-world human interactions alsoinvolve dynamic visual context (similar to videos) as well as dialogueexchanges among multiple speakers.To move closer towards such multimodalconversational skills and visually-situated applications, we introduce a newvideo-context, many-speaker dialogue dataset based on live-broadcast soccergame videos and chats from Twitch.tv.This challenging testbed allows us todevelop visually-grounded dialogue models that should generate relevanttemporal and spatial event language from the live video, while also beingrelevant to the chat history.For strong baselines, we also present severaldiscriminative and generative models, e.g., based on tridirectional attentionflow (TriDAF).We evaluate these models via retrieval ranking-recall, automaticphrase-matching metrics, as well as human evaluation studies.We also presentdataset analyses, model ablations, and visualizations to understand thecontribution of different modalities and model components.论文代码开源:用子词包推广词嵌入(Generalizing Word Embeddings using Bag of Subwords) K6vp9g862j388zpg.jpg
URL地址:https://arxiv.org/abs/1809.04560v1     ----pdf下载地址:https://arxiv.org/pdf/1809.04560v1    ----         ----github下载地址:https://github.com/ramakanth-pasunuru/video-dialogue    ----    论文代码开源:用子词包推广词嵌入(Generalizing Word Embeddings using Bag of Subwords)请注意该人工智能论文代码开源在github,大部分是python写的,框架可能是tensorflow或者pytorch,keras,至于具体是哪一个没有完全测试。
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

admin当前离线
管理员

查看:845 | 回复:0

快速回复 返回顶部 返回列表