copyright | lastupdated | subcollection | ||
---|---|---|---|---|
|
2021-08-20 |
assistant |
{:shortdesc: .shortdesc} {:new_window: target="_blank"} {:external: target="_blank" .external} {:deprecated: .deprecated} {:important: .important} {:note: .note} {:tip: .tip} {:pre: .pre} {:codeblock: .codeblock} {:screen: .screen} {:javascript: .ph data-hd-programlang='javascript'} {:java: .ph data-hd-programlang='java'} {:python: .ph data-hd-programlang='python'} {:swift: .ph data-hd-programlang='swift'}
{: #filter-reference}
The {{site.data.keyword.conversationshort}} service REST API offers powerful log search capabilities through filter queries. You can use the v2 /logs API filter
parameter to search your skill log for events that match a specified query.
The filter
parameter is a cacheable query that limits the results to those matching the specified filter. You can filter on various objects that are part of the JSON response model (for example, the user input text, the detected intents and entities, or the confidence score).
To see examples of filter queries, see Examples.
For more information about the /logs GET
method and its response model, refer to the API Reference{: external}.
{: #filter-reference-syntax}
The following example shows the general form of a filter query:
Location | Query operator | Term |
---|---|---|
request.input.text |
:: |
"IBM Watson" |
- The location identifies the field that you want to filter on (in this example,
request.input.text
). - The query operator, which specifies the type of matching you want to use (fuzzy matching or exact matching).
- The term specifies the expression or value you want to use to evaluate the field for matching. The term can contain literal text and operators, as described in the next section.
Filtering by intent or entity requires slightly different syntax from filtering on other fields. For more information, see Filtering by intent or entity.
Note: The filter query syntax uses some characters that are not allowed in HTTP queries. Make sure that all special characters, including spaces and quotation marks, are URL encoded when sent as part of an HTTP query. For example, the filter response_timestamp<2020-01-01
would be specified as response_timestamp%3C2020-01-01
.
{: #filter-reference-operators}
You can use the following operators in your filter query.
Operator | Description |
---|---|
: |
Fuzzy match query operator. Prefix the query term with : if you want to match any value that contains the query term, or a grammatical variant of the query term. Fuzzy matching is available for user input text, response output text, and entity values. |
:: |
Exact match query operator. Prefix the query term with :: if you want to match only values that exactly equal the query term. |
:! |
Negative fuzzy match query operator. Prefix the query term with :! if you want to match only values that do not contain the query term or a grammatical variant of the query term. |
::! |
Negative exact match query operator. Prefix the query term with ::! if you want to match only values that do not exactly match the query term. |
<= , >= , > , < |
Comparison operators. Prefix the query term with these operators to match based on arithmetic comparison. |
\ |
Escape operator. Use in queries that include control characters that would otherwise be parsed as operators. For example, \!hello would match the text !hello . |
"" |
Literal phrase. Use to encloses a query term that contains spaces or other special characters. No special characters within the quotation marks are parsed by the API. |
~ |
Approximate match. Append this operator followed by a 1 or 2 to the end of the query term to specify the allowed number of single-character differences between the query string and a match in the response object. For example, car~1 would match car , cat , or cars , but it would not match cats . This operator is not valid when filtering on log_id or any date or time field, or with fuzzy matching. |
* |
Wildcard operator. Matches any sequence of zero or more characters. This operator is not valid when filtering on log_id , language , request.context.system.assistant_id , workspace_id , request.context.metadata.deployment , or any date or time field. |
() , [] |
Grouping operators. Use to enclose a logical grouping of multiple expressions using Boolean operators. |
| |
Boolean or operator. |
, |
Boolean and operator. |
{: #filter-reference-intent-entity}
Because of differences in how intents and entities are stored internally, the syntax for filtering on a specific intent or entity is different from the syntax used for other fields in the returned JSON. To specify an intent
or entity
field within an intents
or entities
collection, you must use the :
match operator instead of a dot.
For example, this query matches any logged event where the response includes a detected intent named hello
:
response.output.intents:intent::hello
Note the :
operator in place of a dot (intents:intent)
Use the same pattern to match on any field of a detected intent or entity in the response. For example, this query matches any logged event where the response includes a detected entity with the value soda
:
response.output.entities:value::soda
Similarly, you can filter on intents or entities sent as part of the request, as in this example:
request.input.intents:intent::hello
{: #filter-reference-fields}
To filter on another field in the log data, specify the location as a path identifying the levels of nested objects in the JSON response from the /logs API. Use dots (.
) to specify successive levels of nesting in the JSON data. For example, the location request.input.text
idenfities the user input text field as shown in the following JSON fragment:
"request": {
"input": {
"text": "Good morning"
}
}
Filtering is not available for all fields. You can filter on the following fields:
assistant_id
customer_id
language
request.context.global.system.user_id
request.input.text
request_timestamp
response.context.global.system.user_id
response.output.entities
response.output.intents
response_timestamp
session_id
skill_id
snapshot
Filtering on other fields is not currently supported.
{: #filter-reference-examples}
The following examples illustrate various types of queries using this syntax.
Description | Query |
---|---|
The date of the response is in the month of July 2020. | response_timestamp>=2020-07-01,response_timestamp<2020-08-01 |
The timestamp of the response is earlier than 2019-11-01T04:00:00.000Z . |
response_timestamp<2019-11-01T04:00:00.000Z |
The message is labeled with the customer ID my_id . |
customer_id::my_id |
The message was sent to a specific assistant. | assistant_id::dcd5c5ad-f3a1-4345-89c5-708b0b5ff4f7 |
The user input text contains the word "order" or a grammatical variant (for example, orders or ordering . |
request.input.text:order |
An intent name in the response exactly matches place_order . |
response.output.intents:intent::place_order |
An entity name in the response exactly matches beverage . |
response.output.entities:entity::beverage |
No intent name in the response exactly matches order . |
response.intents:intent::!order |
The user input text does not contain the word "order" or a grammatical variant. | request.input.text:!order |
The user input text contains the string !hello . |
request.input.text:\!hello |
The user input text contains the string IBM Watson . |
request.input.text:"IBM Watson" |
The user input text contains a string that has no more than 2 single-character differences from Watson . |
request.input.text:Watson~2 |
The user input text contains a string consisting of comm , followed by zero or more additional characters, followed by s . |
request.input.text:comm*s |
An intent name in the response exactly matches either hello or goodbye . |
response.output.intents:intent::(hello|goodbye) |
An intent name in the response exactly matches order , and an entity name in the response exactly matches beverage . |
[response.output.intents:intent::order,response.output.entities:entity::beverage] |
If your application is still using the v1 API, you can query and filter logs using the v1 /logs method. The filtering syntax is the same, but the structure of v1 logs and message requests is different. For more information, see API Reference{: external}.
With the v1 /logs API, you can filter on the following fields:
language
meta.message.entities_count
request.context.metadata.deployment
request.context.system.assistant_id
request.input.text
response.context.conversation_id
response.entities
response.input.text
response.intents
response.top_intent
workspace_id
Filtering on other fields is not currently supported.