Master Semantic Networks

Constructing Semantic Networks from Texts

• Semantic networks are graphs that represent words, concepts, multi-grams, or sets of related words.
• Preprocessing methods such as removing punctuation, stopwords, and reducing the number of words through stemming or lemmatization are used to obtain a set of words that express more meaning.
• Networks are built by mapping the connections between words using their co-occurrence within the analyzed texts.
• An edge connects two nodes when the words or concepts they represent appear close to one another in the analyzed texts, and the frequency of co-occurrences gives the weight of this edge.
• A maximum distance between two terms is necessary to define their co-occurrence, and values of 5 or 7 words apart provide stable and robust results.
• Co-occurrences can be defined by segmenting texts into sentences, taking into account punctuation and considering words that occur in the same sentence, but this method has limitations.
• Semantic networks are usually undirected, but studying the direction of co-occurrence relationships can sometimes be useful.
• Many co-occurrences are either spurious or negligible in weight, and edges below a certain weight can be removed.
• Alternative approaches to defining nodes in semantic networks include identifying named entities or considering only bigrams or trigrams, but there are limitations to these approaches.
• Unigram networks can identify bigrams or words that only make sense when combined by using the weight of the edges to evaluate the magnitude of co-occurrence between two words.
• It is important to consider the connections that words have with other words, verbs, or text elements, which may be lost if only bigrams are considered.
• Using excessively high threshold values may result in unrelated words being linked, and common mistakes such as combining words from different text documents should be avoided.

Constructing Semantic Networks from Texts

• Semantic networks are graphs that represent words, concepts, multi-grams, or sets of related words.
• Preprocessing methods such as removing punctuation, stopwords, and reducing the number of words through stemming or lemmatization are used to obtain a set of words that express more meaning.
• Networks are built by mapping the connections between words using their co-occurrence within the analyzed texts.
• An edge connects two nodes when the words or concepts they represent appear close to one another in the analyzed texts, and the frequency of co-occurrences gives the weight of this edge.
• A maximum distance between two terms is necessary to define their co-occurrence, and values of 5 or 7 words apart provide stable and robust results.
• Co-occurrences can be defined by segmenting texts into sentences, taking into account punctuation and considering words that occur in the same sentence, but this method has limitations.
• Semantic networks are usually undirected, but studying the direction of co-occurrence relationships can sometimes be useful.
• Many co-occurrences are either spurious or negligible in weight, and edges below a certain weight can be removed.
• Alternative approaches to defining nodes in semantic networks include identifying named entities or considering only bigrams or trigrams, but there are limitations to these approaches.
• Unigram networks can identify bigrams or words that only make sense when combined by using the weight of the edges to evaluate the magnitude of co-occurrence between two words.
• It is important to consider the connections that words have with other words, verbs, or text elements, which may be lost if only bigrams are considered.
• Using excessively high threshold values may result in unrelated words being linked, and common mistakes such as combining words from different text documents should be avoided.