Table 1 Illustration of instruction data formulation for instruction tuning.
From: Leveraging multimodal large language model for multimodal sequential recommendation
Instruction |
Based on the user’s historical multimodal interaction sequence, please deduce whether the user might like the target item next by responding \(\backslash\)” Yes. \(\backslash\)” or \(\backslash\)” No. \(\backslash\)”.\(\backslash\)n |
Input |
The user_\(\left\{ user\_id \right\}\) is \(\left\{ user\_profile \right\}\) and based on the previous multimodal interaction history ,his preferences are summarized as: \(\backslash\)n |
\(\backslash\)” \(\left\{ user\_preference \right\}\) \(\backslash\)” \(\backslash\)n |
Whether user_ \(\left\{ user\_id \right\}\) |
will like the target item |
\(\backslash\)” \(\left\{ \left( image_{n+1},text_{n+1} \right) \right\}\) \(\backslash\)” |
next \(\backslash\)? \(\backslash\)n |
Response |
\(\backslash\)” Yes. \(\backslash\)” if label = 1 else \(\backslash\)” No. \(\backslash\)” |
