# NLP Models

Controls which Natural Language Processing (NLP) models Aisera runs against user messages. These models analyze the text of each message to extract named entities such as people and organizations, identify the grammatical structure of the query, and evaluate the emotional tone of the conversation. The results are used to improve search accuracy, intent matching, and conversation quality. Disabling models reduces processing overhead but may affect how accurately Aisera understands and responds to user messages.

### NER Model

| **Type**    | Checkbox |
| ----------- | -------- |
| **Default** | Enabled  |

Controls whether Aisera runs Named Entity Recognition (NER) on user messages to identify and extract entities such as people, organizations, dates, and locations. When enabled, Aisera analyzes each user message before processing the request and uses the extracted entities to improve search accuracy, particularly for ontology-based content matching.

Disabling this setting removes entity-based query enrichment and reduces the accuracy of ontology-based search results, but may reduce processing latency.

See also: [Ontology Filters](https://docs.aisera.com/aisera-platform/tenant-setup/aisera-platform-configuration/tenant-configuration-settings/intent-and-search#ontology-filters)

### Sentiment Model

| **Type**    | Checkbox |
| ----------- | -------- |
| **Default** | Disabled |

Controls whether Aisera runs sentiment analysis on user messages during a conversation. When enabled, Aisera evaluates the emotional tone of each user message across a five-point scale from Negative to Positive. If the sentiment is detected as negative or tending negative, Aisera fires a negative sentiment event containing the user's details, initial request, and conversation history up to that point. This event can be used to trigger automated responses such as escalation flows or notifications.

Sentiment results are also stored in the conversation audit record and included in AI-driven ticket resolution predictions when a conversation results in a ticket.

Enable this when you want to automatically detect and respond to users showing signs of frustration, or when you want sentiment data available in audit records and ticket predictions.

### PredObj Extraction Model

| **Type**    | Checkbox |
| ----------- | -------- |
| **Default** | Enabled  |

Controls whether Aisera runs predicate-object extraction on user messages. When enabled, Aisera parses each user message to identify predicate and object structures, which are used to improve search query construction for complex or ambiguous phrasing.

Disabling this setting means predicate extraction relies entirely on Word Annotation model results, which may reduce search accuracy for queries with complex phrasing.

### Word Annotation Model

| **Type**    | Checkbox |
| ----------- | -------- |
| **Default** | Enabled  |

Controls whether Aisera runs part-of-speech tagging on user messages. When enabled, Aisera analyzes each user message to identify verbs, nouns, and other word types, which are used as the primary source for predicate and noun extraction when constructing search queries.

Disabling this setting has a broader impact than disabling other NLP models. Predicate extraction falls back to the [PredObj Extraction Model](#predobj-extraction-model) when this is disabled, but noun extraction has no fallback and produces no results if this model is not running.

### Model Execution Timeout(sec)

| **Type**    | Text field (integers) |
| ----------- | --------------------- |
| **Default** | `30`                  |

No description available.