Knowledge Level Tags: Applied to Collaborative Recommender Systems on the Web

The paper addresses a problem that Twitter researchers face of filtering and identifying relevant tweets by proposing a method to recommend 'better' hashtags to users composing tweet-like content. In doing so, they also propose a number of metrics to determine the quality of a hashtag. They demonstrate their method by means of an implemented system, Cognomy, which is also used to conduct a user experiment to validate their method.

There are multiple problems with the problem statement and the paper. 1. It is not clear how and to what extent better hashtag usage by users will improve the ability of a generic unspecified researcher to identify relevant tweets for their purpose. 2. It is not clear what the characteristics of a 'high-quality' hashtag are and whether users can recognise and use these appropriately when shown a number of examples. 3. There exist multiple hashtag recommendation algorithms in the recommendation systems literature. There is no reference to these algorithms, only to collaborative filtering, so it is not clear how this work relates to existing literature in the relevant field. 4. While there is lengthy discussion of the metrics, there is minimal discussion of how the proposed metrics capture desirable features of high-quality hashtags.

In summary, the paper does not demonstrate or contribute knowledge or experience that is clearly additive to the state of the art. Furthermore, there is no evidence that the proposed recommendation method and metrics address the original problem posed by the paper.

The quality of the writing is strikingly poor with missing words and convoluted language that is hard to comprehend and sometimes sound nonsensical, e.g. 'The proposed metrics aim to improve the quality of metadata used in posts on social networks, offering the user a set of qualified metadata at the level of knowledge where the term used in the metadata has a better understanding of the collective understanding, helping in the content rating process'. Novel ideas such as 'content rating process' above are mentioned without any link to which content is being rated and how. The words metadata, tags, hashtags, terms, knowledge, collective intelligence are used without precise definitions and with low consistency, making it hard to parse text such as 'the metadata term ... has a low level of knowledge compared to the collective intelligence.'

I therefore recommend this paper be rejected in its current state.

The authors have provided resources and data on easily accessible pages. However, I suspect it will be hard to replicate their experiments as I do not see the tweet data and only hashtag data for a sample. Furthermore, the code comments and variable names are in Portuguese (presumably), making it hard to understand the code. There are also similar significant language issues as with the paper.

The paper aims to propose a tag recommendation model for collaborative systems on the web, focusing mainly on the social media platform "X" (formerly known as Twitter). In the paper, the authors stated the following main contributions of the paper: (i) a method of classifying and recommending metadata; (ii) a set of metrics to measure the knowledge level of tags/metadata; (iii) applying visual resources to improve interpretation in the tagging process; and (iv) prototype tool development for evaluation.

The paper proposes three type of metrics: (i) KLE - Knowledge Level Estimate, measuring the level of agreement between user-chosen tags and system-generated tags (based on tags produced by other users in the systems); (ii) KLA - Knowledge Level Adaptation, measuring the level and identify possible deviation of user knowledge about the domain; and (iii) MLK - Metadata Knowledge Level, measuring the added knowledge to the tag/metadata in the search process; sum of KLA + KLE.

It is hard to accept the paper in its current state, mainly due to no apparent contribution to or application of semantic web technologies as part of the proposed approach. The SWJ webpage for authors [1] clearly states, "The journal invites high-quality submissions on all topics related to the Semantic Web, including the use of semantic technologies in other contexts than the World Wide Web", which is not the case with this article.

A GitHub URL for article resources is available, and it contains (i) source code, (ii) example data, and (iii) a README file containing information to replicate the experiment. The resource further clarifies that no semantic web artefacts are involved.

In addition, there are several issues with the paper:

(1) Unclear research gaps and research questions
The topic of tag recommendation (especially on "X"/Twitter) has been investigated for many years. While several approaches to the topic are mentioned and explained in the related work section, there are no apparent research gaps that the authors wanted to address regarding the limitations of the existing approaches.

(2) Limited evaluation and generalization of the approach
The evaluation of the approach is conducted within a specific chosen topic. One may question whether the result will differ if a different topic is selected. For user evaluation (Section 8.2.4; Experiment II), how the evaluation is being conducted needs to be clarified, e.g., Are users tasked to propose their hashtags given a tweet? Are they only use the cognomy tool? Are there any control group that conducted the tasks without using the tool?

Further, since the paper's main topic is tag recommendation, it is expected that the paper reports a comparison of their approach and state of the art regarding the performance and/or user acceptance of tag recommendation as part of their evaluation. While the result of the Cognomy tool (from the paper) is available, there is no indication of how they fare against state-of-the-art approaches.

(3) Quality of writing
The paper contains excessive use of lengthy compound sentences, which makes it difficult to read and understand (e.g., the second sentence of the abstract consists of five lines of text). Furthermore, the article did not define key terms, such as "Knowledge Level Tag" or "Collective Intelligence". Lastly, there are no clear definitions of the terms "hashtag", "tag", and "metadata", which are sometimes used interchangeably.

The article titled “Knowledge Level Tags: Applied to Collaborative Recommender Systems on the
Web”. The main contribution of this article lies in the introduction of an algorithm for classifying metadata at the knowledge level, addressing limitations in the tagging process within collaborative systems on the web. By utilizing metrics that measure collective intelligence, the proposed model aims to enhance the quality and meaningful relationships of tags with objects, ultimately improving user engagement in the collaborative system.

Introduction section need to rewrite with more valuable points about the proposed work. Some sentences are not clear to understand the motivation of the article.

The related work is better to be presented in a table and compare the presented work with the previous work. A comparative table can help to find the gaps of existing work that can be fulfilled by proposed work. Authors are suggested to highlight the limitations of existing approaches.

Section 3 and 6 are very short, it doesn’t make sense to have an individual section for a few lines. Please merge these lines in previous sections.