251 expand preference model

usersimcrs/simulator/agenda_based/agenda_based_simulator.py

@@ -41,8 +41,12 @@ def __init__(
             nlg: NLG module generating textual responses.
             ratings: Historical ratings.
         """
-        super().__init__(id=id, domain=domain, item_collection=item_collection)
-        self._preference_model = preference_model
+        super().__init__(
+            id=id,
+            domain=domain,
+            item_collection=item_collection,
+            preference_model=preference_model,
+        )
         self._interaction_model = interaction_model
         self._interaction_model.initialize_agenda(self.information_need)
         self._nlu = nlu

usersimcrs/simulator/llm/llm_dual_prompt_user_simulator.py

@@ -5,6 +5,8 @@
 the user response.
 """
 
+from typing import Optional
+
 from dialoguekit.core.utterance import Utterance
 from dialoguekit.participant import DialogueParticipant
 from usersimcrs.core.simulation_domain import SimulationDomain
@@ -19,6 +21,7 @@
     UtteranceGenerationPrompt,
 )
 from usersimcrs.simulator.user_simulator import UserSimulator
+from usersimcrs.user_modeling.preference_model import PreferenceModel
 from usersimcrs.user_modeling.persona import Persona
 
 
@@ -32,7 +35,8 @@ def __init__(
         item_type: str,
         task_definition: str = DEFAULT_TASK_DEFINITION,
         stop_definition: str = DEFAULT_STOP_DEFINITION,
-        persona: Persona = None,
+        persona: Optional[Persona] = None,
+        preference_model: Optional[PreferenceModel] = None,
     ) -> None:
         """Initializes the user simulator.
 
@@ -45,14 +49,23 @@ def __init__(
             stop_definition: Definition of the stop task. Defaults to
               DEFAULT_STOP_DEFINITION.
             persona: Persona of the user. Defaults to None.
+            preference_model: Preference model. Defaults to None.
         """
-        super().__init__(id, domain, item_collection)
+        super().__init__(id, domain, item_collection, preference_model)
         self.llm_interface = llm_interface
         self.generation_prompt = UtteranceGenerationPrompt(
-            self.information_need, item_type, task_definition, persona
+            self.information_need,
+            item_type,
+            task_definition,
+            persona,
+            preference_model,
         )
         self.stop_prompt = StopPrompt(
-            self.information_need, item_type, stop_definition, persona
+            self.information_need,
+            item_type,
+            stop_definition,
+            persona,
+            preference_model,
         )
 
     def _generate_response(self, agent_utterance: Utterance) -> Utterance:

usersimcrs/simulator/llm/llm_single_prompt_user_simulator.py

@@ -4,6 +4,8 @@
 model.
 """
 
+from typing import Optional
+
 from dialoguekit.core.utterance import Utterance
 from dialoguekit.participant import DialogueParticipant
 from usersimcrs.core.simulation_domain import SimulationDomain
@@ -14,6 +16,7 @@
     UtteranceGenerationPrompt,
 )
 from usersimcrs.simulator.user_simulator import UserSimulator
+from usersimcrs.user_modeling.preference_model import PreferenceModel
 from usersimcrs.user_modeling.persona import Persona
 
 
@@ -26,7 +29,8 @@ def __init__(
         llm_interface: LLMInterface,
         item_type: str,
         task_definition: str = DEFAULT_TASK_DEFINITION,
-        persona: Persona = None,
+        persona: Optional[Persona] = None,
+        preference_model: Optional[PreferenceModel] = None,
     ) -> None:
         """Initializes the user simulator.
 
@@ -37,11 +41,16 @@ def __init__(
             task_definition: Definition of the task to be performed.
               Defaults to DEFAULT_TASK_DEFINITION.
             persona: Persona of the user. Defaults to None.
+            preference_model: Preference model. Defaults to None.
         """
-        super().__init__(id, domain, item_collection)
+        super().__init__(id, domain, item_collection, preference_model)
         self.llm_interface = llm_interface
         self.prompt = UtteranceGenerationPrompt(
-            self.information_need, item_type, task_definition, persona
+            self.information_need,
+            item_type,
+            task_definition,
+            persona,
+            preference_model,
         )
 
     def _generate_response(self, agent_utterance: Utterance) -> Utterance:

usersimcrs/simulator/llm/prompt/prompt.py

@@ -1,10 +1,12 @@
 """Interface for prompt."""
 
 from abc import ABC, abstractmethod
+from typing import Optional
 
 from dialoguekit.core.utterance import Utterance
 from dialoguekit.participant.participant import DialogueParticipant
 from usersimcrs.core.information_need import InformationNeed
+from usersimcrs.user_modeling.preference_model import PreferenceModel
 from usersimcrs.user_modeling.persona import Persona
 
 
@@ -14,7 +16,8 @@ def __init__(
         information_need: InformationNeed,
         item_type: str,
         prompt_definition: str,
-        persona: Persona = None,
+        persona: Optional[Persona] = None,
+        preference_model: Optional[PreferenceModel] = None,
     ) -> None:
         """Initializes the prompt.
 
@@ -23,18 +26,66 @@ def __init__(
             item_type: The type of the item to be recommended.
             prompt_definition: The definition of the task to be performed.
             persona: The persona of the user. Defaults to None.
+            preference_model: Preference model. Defaults to None.
         """
         self.information_need = information_need
         self.item_type = item_type
         self.prompt_definition = prompt_definition
         self.persona = persona
+        self.preference_model = preference_model
         self._initial_prompt = self.build_new_prompt()
         self._prompt_context = ""
 
     @property
     def prompt_text(self) -> str:
         """Prompt for the user simulator."""
-        return self._initial_prompt + "\n" + self._prompt_context
+        return (
+            self._initial_prompt
+            + "\n"
+            + self._preference_context
+            + "\n"
+            + self._prompt_context
+        )
+
+    @property
+    def _preference_context(self) -> str:
+        """Returns current preference context for the prompt."""
+        if not self.preference_model:
+            return ""
+
+        preference_summary = self.preference_model.get_preference_summary()
+        if not preference_summary:
+            return ""
+
+        return (
+            "PREFERENCES: "
+            f"{preference_summary}\n"
+            f"{self._preference_context_guidance}\n"
+        )
+
+    @property
+    def _preference_context_guidance(self) -> str:
+        """Returns the guidance appended after the preference summary."""
+        return (
+            "Treat these preferences as soft background tastes, not as new "
+            "requirements."
+        )
+
+    @property
+    def _persona_text(self) -> str:
+        """Returns a compact persona description."""
+        if not self.persona:
+            return ""
+
+        if self.persona.persona_description:
+            return self.persona.persona_description
+
+        return ", ".join(
+            [
+                f"{key}={value}"
+                for key, value in self.persona.characteristics.items()
+            ]
+        )
 
     @abstractmethod
     def build_new_prompt(self, **kwargs) -> str:

usersimcrs/simulator/llm/prompt/stop_prompt.py

@@ -1,7 +1,10 @@
 """Define the prompt for stopping the conversation."""
 
+from typing import Optional
+
 from usersimcrs.core.information_need import InformationNeed
 from usersimcrs.simulator.llm.prompt.prompt import Prompt
+from usersimcrs.user_modeling.preference_model import PreferenceModel
 from usersimcrs.user_modeling.persona import Persona
 
 DEFAULT_STOP_DEFINITION = (
@@ -21,7 +24,8 @@ def __init__(
         information_need: InformationNeed,
         item_type: str,
         prompt_definition: str = DEFAULT_STOP_DEFINITION,
-        persona: Persona = None,
+        persona: Optional[Persona] = None,
+        preference_model: Optional[PreferenceModel] = None,
     ) -> None:
         """Initializes the prompt.
 
@@ -31,20 +35,37 @@ def __init__(
             prompt_definition: The definition of the task to be performed.
               Defaults to DEFAULT_STOP_DEFINITION.
             persona: The persona of the user. Defaults to None.
+            preference_model: Preference model. Defaults to None.
         """
         super().__init__(
-            information_need, item_type, prompt_definition, persona
+            information_need,
+            item_type,
+            prompt_definition,
+            persona,
+            preference_model,
         )
 
     @property
     def prompt_text(self) -> str:
         """Prompt for the user simulator."""
+        prompt_text = self._initial_prompt
+
+        if self._preference_context:
+            prompt_text += self._preference_context.rstrip("\n")
+
+        prompt_text += "\n"
+
+        if self._prompt_context:
+            prompt_text += self._prompt_context.rstrip("\n")
+
+        return prompt_text + "\nCONTINUE: "
+
+    @property
+    def _preference_context_guidance(self) -> str:
+        """Returns the guidance appended after the preference summary."""
         return (
-            self._initial_prompt
-            + "\n"
-            + self._prompt_context
-            + "\n"
-            + "CONTINUE: "
+            "Use these preferences only as soft context when deciding "
+            "whether the conversation is still productive."
         )
 
     def build_new_prompt(self) -> str:
@@ -60,13 +81,7 @@ def build_new_prompt(self) -> str:
                 " Take into account your PERSONA when deciding to stop the "
                 "conversation.\n"
             )
-            stringified_characteristics = ", ".join(
-                [
-                    f"{key}={value}"
-                    for key, value in self.persona.characteristics.items()
-                ]
-            )
-            initial_prompt += f"PERSONA: {stringified_characteristics}\n"
+            initial_prompt += f"PERSONA: {self._persona_text}\n"
 
         initial_prompt += "\nHISTORY:\n"
         return initial_prompt

usersimcrs/simulator/llm/prompt/utterance_generation_prompt.py

@@ -9,10 +9,13 @@
 for Task-Oriented Dialog Systems", arXiv 2306.00774.
 """
 
+from typing import Optional
+
 from dialoguekit.core.utterance import Utterance
 from dialoguekit.participant.participant import DialogueParticipant
 from usersimcrs.core.information_need import InformationNeed
 from usersimcrs.simulator.llm.prompt.prompt import Prompt
+from usersimcrs.user_modeling.preference_model import PreferenceModel
 from usersimcrs.user_modeling.persona import Persona
 
 DEFAULT_TASK_DEFINITION = (
@@ -34,7 +37,8 @@ def __init__(
         information_need: InformationNeed,
         item_type: str,
         prompt_definition: str = DEFAULT_TASK_DEFINITION,
-        persona: Persona = None,
+        persona: Optional[Persona] = None,
+        preference_model: Optional[PreferenceModel] = None,
     ) -> None:
         """Initializes the prompt.
 
@@ -44,9 +48,14 @@ def __init__(
             prompt_definition: The definition of the task to be performed.
               Defaults to DEFAULT_TASK_DEFINITION.
             persona: The persona of the user. Defaults to None.
+            preference_model: Preference model. Defaults to None.
         """
         super().__init__(
-            information_need, item_type, prompt_definition, persona
+            information_need,
+            item_type,
+            prompt_definition,
+            persona,
+            preference_model,
         )
 
     def build_new_prompt(self) -> str:
@@ -61,13 +70,7 @@ def build_new_prompt(self) -> str:
             initial_prompt += (
                 " Adapt your responses considering your PERSONA.\n"
             )
-            stringified_characteristics = ", ".join(
-                [
-                    f"{key}={value}"
-                    for key, value in self.persona.characteristics.items()
-                ]
-            )
-            initial_prompt += f"PERSONA: {stringified_characteristics}\n"
+            initial_prompt += f"PERSONA: {self._persona_text}\n"
         else:
             initial_prompt += (
                 "Be precise with the REQUIREMENTS, clear and concise.\n"
@@ -90,6 +93,14 @@ def build_new_prompt(self) -> str:
         )
         return initial_prompt
 
+    @property
+    def _preference_context_guidance(self) -> str:
+        """Returns the guidance appended after the preference summary."""
+        return (
+            "Use these preferences to shape what the user accepts, rejects, "
+            "asks to avoid, or follows up on."
+        )
+
     def update_prompt_context(
         self, utterance: Utterance, participant: DialogueParticipant
     ) -> None:

usersimcrs/simulator/user_simulator.py

@@ -1,13 +1,15 @@
 """User simulator abstract class."""
 
 from abc import ABC, abstractmethod
+from typing import Optional
 
 from dialoguekit.core.annotated_utterance import AnnotatedUtterance
 from dialoguekit.core.utterance import Utterance
 from dialoguekit.participant.user import User, UserType
 from usersimcrs.core.information_need import generate_random_information_need
 from usersimcrs.core.simulation_domain import SimulationDomain
 from usersimcrs.items.item_collection import ItemCollection
+from usersimcrs.user_modeling.preference_model import PreferenceModel
 
 
 class UserSimulator(User, ABC):
@@ -16,11 +18,13 @@ def __init__(
         id: str,
         domain: SimulationDomain,
         item_collection: ItemCollection,
+        preference_model: Optional[PreferenceModel] = None,
     ) -> None:
         """Initializes the user simulator."""
         super().__init__(id, UserType.SIMULATOR)
         self._domain = domain
         self._item_collection = item_collection
+        self.preference_model = preference_model
         self.get_new_information_need()
 
     def get_new_information_need(self) -> None:
@@ -53,4 +57,6 @@ def receive_utterance(self, utterance: Utterance) -> None:
         response = self._generate_response(utterance)
         if not isinstance(response, AnnotatedUtterance):
             response = AnnotatedUtterance.from_utterance(response)
+        if self.preference_model:
+            self.preference_model.update_from_utterance(response)
         self._dialogue_connector.register_user_utterance(response)