Search at Slack

https://slack.engineering/search-at-slack-431f8c80619e

Slack搜索按照两个指标排序: Relevance(相关性)和Recent(时间)

Slack的检索和Google的检索在数据方面差异很大：

• web search会出现一些热门query, 而这点slack search里面不太可能出现
• slack 对用户的work graph以及search context更加了解
• slack search不需要考虑SEO game出现的spam问题
• 数据量相比web search小很多，所以可以将更多的计算资源放在rank上
• 在展现搜索结果上也具有很高的灵活性，比如可以展现对话上下文（比如直接返回可以回答这个问题的人）

recent的实现相对比较直接一些，而relevance则涉及到了ranking问题。在我们优化之前，relevance只占用了17%的搜索流量，剩下的都是recent, 并且从头部位置的ctr来看，relevance还有很大的改进空间。一种改进的办法就是利用用户的work graph，说白了就是用户行为和社交关系。优化之后，relevance的流量提升了9%, 然后position 1的ctr增加了27%.

Learning to Rank是个很关键的改进，slack search的做法是首先使用solr/lucene的基本功能从索引里面把相关性的比较高的docs筛选出来，然后在application space上做一些计算密集型的ltr工作.

To achieve this we settled on a two-stage approach: we would leverage Solr’s custom sorting functionality to retrieve a set of messages ranked by only the select few features that were easy for Solr to compute, and then re-rank those messages in the application layer according to the full set of features, weighted appropriately.

根据点击的行为反馈，可以对每个pair of docs生成0/1这样的结果，然后使用这些数据作为训练数据来构建模型。

点击行为会有click-position bias的问题，比如position n肯定比如position n+1曝光次数多，ctr也高出30%（这个是slack给出他们的数据）。所以在采样数据的时候，会对排在后面的doc进行点击的过采样，来修正这个偏差。现在生成这种pair的粒度比如是20 docs per page的话，有可能后面docs根本没有出现，但是我们也会将这些hidden docs记录到pairs里面，好像这种问题没有特别好的办法，否则需要在前端反馈哪些docs展现了。

训练算法和features如下：

Using this dataset, we trained a model using SparkML’s built-in SVM algorithm. The model determined that the following signals were the most significant:

• The age of the message
• The Lucene score of the message with respect to the query
• The searcher’s affinity to the author of the message (we defined affinity of one user for another as the propensity of that user to read the other’s messages — a subject for another post!)
• The priority score of the searcher’s DM channel with the message author
• The searcher’s priority score for the channel the message appeared in
• Whether the message author is the same as the searcher
• Whether the message was pinned, starred or had emoji reactions
• The propensity of searchers to click on other messages from the channel the message appeared in
• Aspects of the content of the message, such as word count, presence of line breaks, emoji and formatting.

Notably, aside from the Lucene “match” score, we have not yet incorporated any other semantic features of the message itself in our models.

为了能更好地提升搜索结果，在默认的"Recent"结果里面增加了"Top Results"这个版块。

• 在搜索Recent的时候，同时会去请求Relevance
• 根据Relevance结果的diversity/quantity来决定是否要出现Top Results这个版块
• 如果出现的话，把Relevance结果的头三个显示在里面