LUIS中能否合并两个datetimeV2实体？实现单一日期实体识别

Great question! I’ve run into this exact scenario before where LUIS’s built-in datetimeV2 entity splits combined relative+absolute date references into separate entities. Here’s how you can get it to recognize "today 26th march" as a single logical entity:

Option 1: Use a Composite Entity #

This is the most straightforward approach to group the two datetimeV2.date instances into one cohesive entity:

• Head to your LUIS model, create a new Composite Entity (e.g., CombinedDate), and add builtin.datetimeV2.date as its child entity.
• Add training utterances where you explicitly label the entire phrase "today 26th march" as the CombinedDate composite entity, while also marking "today" and "26th march" as child datetimeV2.date entities within it.
• Repeat with similar examples like "this day 15th june", "now 30th november" to teach LUIS that these paired date references belong together. LUIS will gradually learn to associate the relative and absolute date terms as part of a single entity.

Option 2: Leverage Patterns to Enforce Merging #

If you want more control over specific phrase structures, use Patterns:

• Create a Pattern tied to your target intent, like:

  [{CombinedDate:today {builtin.datetimeV2.date}}]
  

• For broader coverage, use a more flexible pattern that matches any two datetimeV2.date terms in sequence:

  [{CombinedDate:{builtin.datetimeV2.date} {builtin.datetimeV2.date}}]
  

• This tells LUIS to treat consecutive date references as part of the same CombinedDate entity when they fit the pattern.

Option 3: Post-Processing Logic (Fallback) #

If the above methods don’t cover all edge cases, add a quick check in your application code:

• When LUIS returns two datetimeV2.date entities, inspect their resolution values. If both resolve to the same actual date (e.g., "today" maps to 2024-03-26 and "26th march" also maps to 2024-03-26), merge them into a single entity in your code before using the data.
• This works especially well if you’re dealing with users who repeat the same date for emphasis.

Key Notes #

• Make sure you have enough diverse training samples—LUIS relies on examples to learn these associations.
• Test thoroughly with variations of the phrase to ensure the model behaves as expected.

内容的提问来源于stack exchange，提问作者Ferdinand Fejskid