Future of YouTube Comments Powered By Artificial Intelligence

YouTube评论的未来由人工智能提供支持

Google introduces its SmartReply feature for YouTube creators. This technology with the help of artificial intelligence analyses the message from the user and predicts the response for it.

Google为YouTube创作者引入了SmartReply功能。 这项技术借助人工智能来分析来自用户的消息并预测其响应。

Initially, this A.I. based technology was used in Gmail, android messages, and other android devices to suggest messages and replies.

最初，这种基于AI的技术被用于Gmail，Android消息和其他android设备中，以建议消息和回复。

This feature is now implemented in the YouTube studio. From YouTube Studio’s comments section, creators can filter, view and respond to comments from across their channels. Currently, SmartReply for YouTube is only available for English and Spanish language creators.

现在已在YouTube工作室中实现了此功能。 在YouTube Studio的评论部分，创作者可以过滤，查看和回复来自各个渠道的评论。 目前，适用于YouTube的SmartReply仅适用于英语和西班牙语的创作者。

As YouTube developers receive an enormous volume of responses to their videos from viewers, SmartReply will save a great amount of time and efforts in replying to each comment manually and allowing them to be in touch with the viewers more easily. However, YouTube replies will be still chosen by the creator from the suggestions provided by the software. This technology will not post anything by itself but will only provide suggestions.

由于YouTube开发人员收到了观众对视频的大量回复，因此SmartReply将节省大量的时间和精力，以便手动回复每个评论，并使他们更轻松地与观众保持联系。 但是，创建者仍会从该软件提供的建议中选择YouTube答复。 这项技术本身不会发布任何内容，只会提供建议。

According to Google,” In comparison to emails, which tend to be long and dominated by formal language, YouTube comments reveal complex patterns of language switching, abbreviated words, slang, inconsistent usage of punctuation, and heavy utilization of emoji.” This was the difficulty faced by Google while implementing the SmartReply feature on YouTube.

谷歌认为，“ 与电子邮件相比，电子邮件通常较长且受正式语言支配，YouTube评论显示出语言切换，缩写词，语，标点符号用法不一致以及表情符号使用率高的复杂模式 。” 这就是Google在YouTube上实现SmartReply功能时面临的困难。

(Image source: Google)

(图片来源： Google)

Initially, SmartReply encoded word-by-word with a neural network and then decoded prospective replies with another neural network. But this method was computationally expensive, so in place of this approach, a system was developed where it searches through a predefined list of suggestions for the most perfect replies. This system worked well for Gmail and other SmartReply systems but it faced challenges for YouTube (because of its content like slang, abbreviated words, emoji, ASCII, typos, etc.), to accomplish this Google applied a deep transform network which is able to model words and phrases from the sequence of characters or bytes and making SmartReply’s first cross-lingual and character byte-based version of the technology.

最初，SmartReply使用神经网络逐字编码，然后使用另一个神经网络对预期答复进行解码。 但是此方法在计算上很昂贵，因此开发了一个系统来代替此方法，在该系统中搜索预定义的建议列表以获取最完美的答复。 这个系统在Gmail和其他SmartReply系统上运作良好，但在YouTube上却面临挑战(因为它的内容包括like语，缩写词，表情符号，ASCII，错别字等)，以实现此Google应用了深度转换网络，该网络能够从字符或字节的序列中模拟单词和短语，并使其成为SmartReply的第一个跨语言和基于字符字节的技术版本。

Ideally, they want that the software to provide suggestions only when the creator would reply to the comment and when the model has a high chance of providing a sensible and specific response or it is very likely to be useful to the viewer’s comment.

理想情况下，他们希望该软件仅在创建者会回复评论时以及模型有很大机会提供明智而具体的响应或非常有可能对查看者的评论有用时才提供建议。

Originally published at https://patataeater.blogspot.com on July 13, 2020.

最初于 2020年7月13日 发布在 https://patataeater.blogspot.com 上。