Applying Network Science to Semantic Text Analysis

Several different research fields deal with text, such as text mining, computational linguistics, machine learning, information retrieval, semantic web and crowdsourcing. Grobelnik states the importance of an integration of these research areas in order to reach a complete solution to the problem of text understanding. The novel analysis methods proposed in a paper by Livia Celardo et al. focused on experimenting with cluster analysis of the semantic network. We adjusted our network analysis process significantly throughout the project, so Celardo et al.’s work on improving analysis accuracy related to our struggles with creating realistic keyword clusters from our network. Celardo et al. aimed to improve analysis accuracy by modeling data more realistically with the incorporation of text co-clusters. Whereas current models often create network clusters where the mean value converges toward the cluster center, these researchers expanded the text clustering methods by partitioning both the rows and columns in the matrix of similarities.

There are important initiatives to the development of researches for other languages, as an example, we have the ACM Transactions on Asian and Low-Resource Language Information Processing , an ACM journal specific for that subject. Papers expanding existing text analysis methods or inventing new methods often shed light on existing issues in the field of network science text analysis, which we found very helpful in assessing the pros and cons of our method choices. Two such research papers we found focused on training and analyzing new neural network models to rank similarities of texts, as a more versatile method than existing work. In a paper by Kiran Mysore Ravi et al., they trained a Long Short Term Memory variation on an RNN model to analyze unprocessed raw text, which allowed them to analyze diverse text datasets with a central method. Similarly, in a paper by Chanzheng Fu et al., the researchers evaluated their new memory neural network model, which outperformed an existing neural network variation. However, whereas Ravi et al. used n-grams to rank similarity in the text, Fu et al. deviate from the n-grams method, which they believe is becoming less relevant as network science methods improve.

Share this article

We can note that text semantics has been addressed more frequently in the last years, when a higher number of text mining studies showed some interest in text semantics. The lower number of studies in the year 2016 can be assigned to the fact that the last searches were conducted in February 2016. However, there is a lack of studies that integrate the different branches of research performed to incorporate text semantics in the text mining process.

With the runtime issue partially resolved, we examined how to translate the kernel matrix into an adjacency matrix. Foxworthy used a cutoff value, where he put an edge between texts with a lower hamming similarity value than the cutoff. Since hamming distance counts the differences, two vectorized strings that are identical will have a hamming distance of 0.

Meaning Representation

Firstly, Kitchenham and Charters state that the systematic review should be performed by two or more researchers. Although our mapping study was planned by two researchers, the study selection and the information extraction phases were conducted by only one due to the resource constraints. In this process, the other researchers reviewed the execution of each systematic mapping phase and their results. Secondly, systematic reviews usually are done based on primary studies only, nevertheless we have also accepted secondary studies as we want an overview of all publications related to the theme.

The adjacency matrix corresponded to a semantic network from which Foxworthy extracted communities and sentiment keywords to characterize the communities. Researchers also often applied common network analysis techniques to their text datasets and semantic networks to discover complex categorizations of the texts. For example, many research papers we read relied on relating data sets to thesauri ontologies to determine similarities and edges in the network. In a paper by Roberto Willrich et al., they performed this type of knowledge base analysis to determine students’ reading comprehension of the text, which is a type of sentiment analysis. Similarly, in a paper by Manuel W Bickel, the researchers used text mining on large climate action plans, and related the resulting data set to three knowledge bases to analyze climate action plans by known methods.

Share this paper

As an example, in the pre-processing step, the user can provide additional information to define a stoplist and support feature selection. In the pattern extraction step, user’s participation can be required when applying a semi-supervised approach. In the post-processing step, the user can evaluate the results according to the expected knowledge usage. The use of Wikipedia is followed by the use of the Chinese-English knowledge database HowNet .

The algorithm is chosen based on the data available and the type of pattern that is expected. If this knowledge meets the process objectives, it can be put available to the users, starting the final step of the process, the knowledge usage. Otherwise, another cycle must be performed, making changes in the data preparation activities and/or in pattern extraction parameters. If any changes in the stated objectives or selected text collection must be made, the text mining process should be restarted at the problem identification step.

Text representation models

In the experiment, three thesauri described categories, then the researchers ranked these categories by their perceived network importance. This type of analysis is very similar to our experiments, since the researchers categorized sentiments in the climate action plans. An ontology also played a key role in this paper, when they translated a vector space model of “document-section-termmatrices” into “document-category-term-matrices” through relations to the ontological categories. Therefore, this paper showed the importance of matrices and models to determine links in a text analysis network.

Stavrianou et al. also present the relation between ontologies and text mining. Ontologies can be used as background semantic text analysis in a text mining process, and the text mining techniques can be used to generate and update ontologies. In simple words, we can say that lexical semantics represents the relationship between lexical items, the meaning of sentences, and the syntax of the sentence. Powerful semantic-enhanced machine learning tools will deliver valuable insights that drive better decision-making and improve customer experience.

Analytics Vidhya App for the Latest blog/Article

This way, we could choose cutoffs that were higher on the scatter-plot and further the intuitive sense that a high hamming value means high similarity. This chapter describes a generic semantic grammar that can be used to encode themes and theme relations in every clause within randomly sampled texts. In a semantic text analysis, the researcher encodes only those parts of the text that fit into the syntactic components of the semantic grammar being applied.

Gain a deeper understanding of the relationships between products and your consumers’ intent. Implement a Connected Inventory of enterprise data assets, based on a knowledge graph, to get business insights about the current status and trends, risk and opportunities, based on a holistic interrelated view of all enterprise assets. In other words, semantic search is an entirely different approach when it compared to the more common keyword-based search approach that relies on matching keywords in the user’s query to the search results to find relevant results. However, with the semantic search approach, there is next level of search relevancy that is never possible with the keyword-based technique. Some studies accepted in this systematic mapping are cited along the presentation of our mapping.

• In this semantic space, alternative forms expressing the same concept are projected to a common representation.
• Consequently, in order to improve text mining results, many text mining researches claim that their solutions treat or consider text semantics in some way.
• We included this research because of its innovative use of the matrix for text analysis, and because they focused on mirroring patterns in real text data.
• However, at this point we had concerns about runtime, since our data set was very large and we were beginning to work on large matrix and network manipulations in the method.
• The author also discusses the generation of background knowledge, which can support reasoning tasks.
• The difficulty inherent to the evaluation of a method based on user’s interaction is a probable reason for the lack of studies considering this approach.