ElasticSearch 映射折叠/对分组文档执行操作的结果

Question

有一个对话列表，每个对话都有一个消息列表。每条消息都有不同的字段和 action field 。我们需要考虑在对话的第一条消息中使用了操作A。 ，在几条消息之后，使用了操作 A.1过了一会儿A.1.1等等(有一个聊天机器人意图列表)。

将对话的消息操作分组将类似于:A > A > A > A.1 > A > A.1 > A.1.1 ...
问题:

我需要使用 ElasticSearch 创建一个报告，该报告将返回 actions group每一次谈话；接下来，我需要对类似的 actions groups 进行分组添加计数；最终将导致 Map<actionsGroup, count>如'A > A.1 > A > A.1 > A.1.1', 3 .

构建 actions group我需要消除每一组重复项；而不是 A > A > A > A.1 > A > A.1 > A.1.1我需要 A > A.1 > A > A.1 > A.1.1 .

我开始做的步骤 :

{
   "collapse":{
      "field":"context.conversationId",
      "inner_hits":{
         "name":"logs",
         "size": 10000,
         "sort":[
            {
               "@timestamp":"asc"
            }
         ]
      }
   },
   "aggs":{
   },
}

我接下来需要什么:

我需要将崩溃的结果映射到单个结果中，例如 A > A.1 > A > A.1 > A.1.1 .我在案例中看到过，或aggr可以使用scripts通过结果，可以创建我需要的操作列表，但是 aggr正在对所有消息进行操作，而不仅仅是对我崩溃的分组消息。有可能使用aggr内部崩溃或类似的解决方案？

我需要对所有折叠的结果值( A > A.1 > A > A.1 > A.1.1 )进行分组，添加计数并得到 Map<actionsGroup, count> .

或:

按 conversationId 对对话消息进行分组字段使用 aggr (我不知道我该怎么做)

使用脚本迭代所有值并创建 actions group对于每一次谈话。 (不确定这是否可能)

使用另一个 aggr对所有值进行分组，返回 Map<actionsGroup, count> .

更新 2:我设法获得了部分结果，但仍然存在一个问题。请查看 here我还需要修复什么。

更新 1:添加一些额外的细节

映射:

"mappings":{
  "properties":{
     "@timestamp":{
        "type":"date",
        "format": "epoch_millis"
     }
     "context":{
        "properties":{
           "action":{
              "type":"keyword"
           },
           "conversationId":{
              "type":"keyword"
           }
        }
     }
  }
}

对话文件样本:

Conversation 1.
{
    "@timestamp": 1579632745000,
    "context": {
        "action": "A",
        "conversationId": "conv_id1",
    }
},
{
    "@timestamp": 1579632745001,
    "context": {
        "action": "A.1",
        "conversationId": "conv_id1",
    }
},
{
    "@timestamp": 1579632745002,
    "context": {
        "action": "A.1.1",
        "conversationId": "conv_id1",
    }
}

Conversation 2.
{
    "@timestamp": 1579632745000,
    "context": {
        "action": "A",
        "conversationId": "conv_id2",
    }
},
{
    "@timestamp": 1579632745001,
    "context": {
        "action": "A.1",
        "conversationId": "conv_id2",
    }
},
{
    "@timestamp": 1579632745002,
    "context": {
        "action": "A.1.1",
        "conversationId": "conv_id2",
    }
}

Conversation 3.
{
    "@timestamp": 1579632745000,
    "context": {
        "action": "B",
        "conversationId": "conv_id3",
    }
},
{
    "@timestamp": 1579632745001,
    "context": {
        "action": "B.1",
        "conversationId": "conv_id3",
    }
}

预期结果:

{
    "A -> A.1 -> A.1.1": 2,
    "B -> B.1": 1
}
Something similar, having this or any other format.

由于我是 elasticsearch 新手，因此每个提示都非常受欢迎。

Answer 1

我使用 scripted_metric 解决了它的弹性。此外，index从初始状态改变。

剧本:

{
   "size": 0,
   "aggs": {
        "intentPathsCountAgg": {
            "scripted_metric": {
                "init_script": "state.messagesList = new ArrayList();",
                "map_script": "long currentMessageTime = doc['messageReceivedEvent.context.timestamp'].value.millis; Map currentMessage = ['conversationId': doc['messageReceivedEvent.context.conversationId.keyword'], 'time': currentMessageTime, 'intentsPath': doc['brainQueryRequestEvent.brainQueryRequest.user_data.intentsHistoryPath.keyword'].value]; state.messagesList.add(currentMessage);",  
                "combine_script": "return state",
                "reduce_script": "List messages = new ArrayList(); Map conversationsMap = new HashMap(); Map intentsMap = new HashMap(); String[] ifElseWorkaround = new String[1]; for (state in states) { messages.addAll(state.messagesList);} messages.stream().forEach((message) -> { Map existingMessage = conversationsMap.get(message.conversationId); if(existingMessage == null || message.time > existingMessage.time) { conversationsMap.put(message.conversationId, ['time': message.time, 'intentsPath': message.intentsPath]); } else { ifElseWorkaround[0] = ''; } }); conversationsMap.entrySet().forEach(conversation -> { if (intentsMap.containsKey(conversation.getValue().intentsPath)) { long intentsCount = intentsMap.get(conversation.getValue().intentsPath) + 1; intentsMap.put(conversation.getValue().intentsPath, intentsCount); } else {intentsMap.put(conversation.getValue().intentsPath, 1L);} }); return intentsMap.entrySet().stream().map(intentPath -> [intentPath.getKey().toString(): intentPath.getValue()]).collect(Collectors.toSet()) "
            }
        }
    }
}

格式化脚本(为了更好的可读性 - 使用 .ts):

scripted_metric: {
  init_script: 'state.messagesList = new ArrayList();',
  map_script: `
    long currentMessageTime = doc['messageReceivedEvent.context.timestamp'].value.millis;
    Map currentMessage = [
      'conversationId': doc['messageReceivedEvent.context.conversationId.keyword'],
      'time': currentMessageTime,
      'intentsPath': doc['brainQueryRequestEvent.brainQueryRequest.user_data.intentsHistoryPath.keyword'].value
    ];
    state.messagesList.add(currentMessage);`,
  combine_script: 'return state',
  reduce_script: `
    List messages = new ArrayList();
    Map conversationsMap = new HashMap();
    Map intentsMap = new HashMap();
    boolean[] ifElseWorkaround = new boolean[1];

    for (state in states) {
      messages.addAll(state.messagesList);
    }

    messages.stream().forEach(message -> {
      Map existingMessage = conversationsMap.get(message.conversationId);
      if(existingMessage == null || message.time > existingMessage.time) {
        conversationsMap.put(message.conversationId, ['time': message.time, 'intentsPath': message.intentsPath]);
      } else {
        ifElseWorkaround[0] = true;
      }
    });

    conversationsMap.entrySet().forEach(conversation -> {
      if (intentsMap.containsKey(conversation.getValue().intentsPath)) {
        long intentsCount = intentsMap.get(conversation.getValue().intentsPath) + 1;
        intentsMap.put(conversation.getValue().intentsPath, intentsCount);
      } else {
        intentsMap.put(conversation.getValue().intentsPath, 1L);
      }
    });

    return intentsMap.entrySet().stream().map(intentPath -> [
      'path': intentPath.getKey().toString(),
      'count': intentPath.getValue()
    ]).collect(Collectors.toSet())`

答案:

{
    "took": 2,
    "timed_out": false,
    "_shards": {
        "total": 5,
        "successful": 5,
        "skipped": 0,
        "failed": 0
    },
    "hits": {
        "total": {
            "value": 11,
            "relation": "eq"
        },
        "max_score": null,
        "hits": []
    },
    "aggregations": {
        "intentPathsCountAgg": {
            "value": [
                {
                    "smallTalk.greet -> smallTalk.greet2 -> smallTalk.greet3": 2
                },
                {
                    "smallTalk.greet -> smallTalk.greet2 -> smallTalk.greet3  -> smallTalk.greet4": 1
                },
                {
                    "smallTalk.greet -> smallTalk.greet2": 1
                }
            ]
        }
    }
}