SMA

CRFs help in situations where the prediction for one element in a sequence depends on the context of the other elements. They are particularly useful for structured prediction problems where the output has some inherent structure or relationships, and the goal is to model these dependencies to improve prediction accuracy.

They are used to model the dependencies between input features and output labels. They are the building blocks that enable CRFs to capture complex patterns and relationships in structured prediction tasks. The combination of feature functions and their associated weights forms the basis for making predictions in a CRF.

"Barack Obama was born in Hawaii."

In this sentence, "Barack Obama" is a named entity, and Named Entity Linking would involve finding the unique identifier associated with the entry for "Barack Obama" in a knowledge base.

Disambiguation works through the use of external lexical resources such as dictionaries and thesauri.

Machine Readable Dictionaries (MRD): For each word in the language vocabulary, a MRD provides:

• A list of meanings
• Definitions (for all word meanings)
• Typical usage examples (for most word meanings)

Thesaurus: adds explicit synonymy relation between word meanings. E.g.:

• plant, works, industrial plant
• plant, flora, plant life

Semantic networks: adds more semantic relations:

• {plant, flora, plant life}
  • hypernym: {organism, being}
  • hypomym: {house plant}, {fungus}, ...
  • meronym: {plant tissue}, {plant part}
  • holonym: {Plantae, kingdom Plantae, plant kingdom} 

The Lesk algorithm

Goal: compare the contexts of ambiguous words (or entities) with the definitions or contexts of candidate senses from a lexical resource (like a dictionary or a knowledge base) --> definition overlap

1. Retrieve from MRD all sense definitions of the words to be disambiguated.

2. Determine the definition overlap for all possible sense combinations.

3. Choose senses that lead to highest overlap.

Example:

We want to understand what is meant by "pine cone" (which is part of a sentence)

We collect all the definitions for both words and see where there is the most overlap (see picture)

The simplified lesk algorithm

Instead of looking for the most overlap between the sense combinations, we look at where there is the most overlap between a definition and the words of the sentence we are looking at (see image)

It belongs to Co-occurrence graphs algorithms of WSI

Proxy has all our requested pages, since we request all pages using the same proxy

Third-party cookies are generated and placed on the user's device by a different website other than the one the user is visiting. They are created when a user visits a website that includes elements from other sites, such as third-party images or ads.