I developed a process to extract your citation and I obtain 52 references in the set we are using, your information is very valuable.
Thanks a lot.
1 Like
Yes please, first we’ll assitance labelling papers to train the AI paper recognition algorithm. After we’ve finalized our list of high-confidence papers there re many ways forward. One option is then making a knowledge graph to help us analyse connections between the papers. Do you have experience with NEO4J knowledge graphs? What do you think of this idea and potential software? https://llm-graph-builder.neo4jlabs.com/
2 Likes
Yes please, first we’ll assistance labelling papers to train the AI paper recognition algorithm, Angel has a plan and software tool to help coming soon. After we’ve finalized our list of high-confidence papers there re many ways forward. One option is then making a knowledge graph to help us analyse connections between the papers. Do you have experience with NEO4J knowledge graphs? What do you think of this idea and potential software? https://llm-graph-builder.neo4jlabs.com/
1 Like
Yes please, first we’ll assist in labelling papers to train the AI paper recognition algorithm, Angel has a plan and software tool to help come soon. After we’ve finalized our list of high-confidence papers there are many ways forward. One option is to make a knowledge graph to help us analyse connections between the papers. Do you have experience with NEO4J knowledge graphs? What do you think of this idea and potential software? https://llm-graph-builder.neo4jlabs.com/
1 Like
Yes please, first we’ll need assistance in labelling papers to train the AI paper recognition algorithm, Angel has a plan and software tool to help come soon. After we’ve finalized our list of high-confidence papers there are many ways forward, so open to suggestions. One option is to make a knowledge graph to help us analyse connections between the papers. Do you have experience with NEO4J knowledge graphs? What do you think of this idea and potential software? https://llm-graph-builder.neo4jlabs.com/
1 Like
Yes please, first we’ll assist in labelling papers to train the AI paper recognition algorithm, Angel has a plan and software tool to help come soon. After we’ve finalized our list of high-confidence papers there are many ways forward. One option is to make a knowledge graph to help us analyse connections between the papers, we’re also open to suggestions. Do you have experience with NEO4J knowledge graphs? What do you think of this idea and potential software? https://llm-graph-builder.neo4jlabs.com/
1 Like
The simplest method is to relate the papers via references, the problem is that to do that effectively every paper must be open or show at least the references.
There is another more philosophical topic, in fact all the papers are relevant because they can bring us inspiration from topics that are not related, but what we can do without discarding is to make a search engine biased to the criteria we use for relevance, so we do not discard nothing, but if we prioritize the most relevant, everything depends on the objectives, fortunately 17,000 are very few, in terms of required infrastructure, if the number of papers, users and queries increases considerably, it can change, I can host the use of members of AWG and this biased approach to PlantAWG, but heavy traffic requires more infrastructure.
I can easily develop something similar to Google academics but very specialized for PlantAWG if you are interested. In all the cases human labeling is key for the quality. After this first approach, the training would be automatic with use, because the clicks of the users on the searches would automatically train the search engine.
1 Like
That would be awesome! 
We clearly have some volunteers in this conversation who seem interested in labelling some of the papers.
2 Likes
Perfect, I can create accounts and send it to volunteers, I can administrate the search engine too, with use the search engine will improve considerably. I have to optimize and analize the performance, maybe the search engine could be ready today.
1 Like
This is really a wonderful idea. I would like to work on it. It would be great, If we can have scifinder data as well. And I feel after 2-3 screenings, human annotation is essential.
2 Likes
This is the first version of the search engine:
https://7os.us/nasa/awg/microgravityanalogueliteraturereview/search/
I’m going to improve the search engine after develop some tools this week, but you can use it now.
1 Like
Note that NASA cyber security setting prevent some people from accessing the site. Solved! Well done 
2 Likes
ok, try with this domain:
https://www.grupoalianzaempresarial.com/nasa/awg/microgravityanalogueliteraturereview/search/
2 Likes
This app has a smooth front end! I think the RPM algorithm might need some work as i noticed a there were a few false positives on the first page… Is it possible to down the complete lists of true positives (for plotting graphs)? It looks great and the 2D clinostat list looked precise, but I’m not sure i understand what’s going on inside this “microgravity analogue literature review software”? Another tangent would be trying to extract the type of microgravity simulation into a column (& potentially another column for the species &/or tissue type)… A lot of potential here! But how best to coordinate the paper labelling / AI training and how to create enough methodological transpancy to ensure user confidence in the results?
I will work on this through search analysis, which requires several users to use the system to facilitate improvements and review patterns. The functionality of the search engine will undergo significant changes this week. Soon, logged-in users will be able to rate searches to modify the ranking. While the relevance parameter is a good starting point for labeling information, we need more labeled content. Although the processes run in the background, they are not continuous; I might refresh the ranking once a day. Today, I will activate the ranking feature. Initially, we may obtain less relevant results, but this week I will develop a more sophisticated ranking method, depending on the characteristics of the data.
You can download the CSV files with positives and false positives (human) and positives and false positives (machine) from the system. The buttons are at the bottom. The machine approach is obviously less precise but aims to reduce human workload. I refresh the machine selection due to the corrections.
Regarding transparency, I make hundreds of changes daily and prefer to publish the methods once finalized. Currently, the search engine is not operating with the methods it will use by the end of the week. However, using the search engine now will help improve it. I have developed a better way to pull information from NIH, but it is a slow process that I have not yet begun. This change will eventually provide advanced search options.
For tissues and species, if you cannot find the information at this moment, it may remain unavailable in the future due to our reliance on titles and abstracts. There are several sources of information, some of which are closed, preventing us from obtaining additional data. It is possible to access more information, but it requires a significant amount of work.
2 Likes
Great work, rapid progress, like the early prototypes… No need to rush
2 Likes
Hi, I was reviewing the contents in the app. It’s just an idea, if we can try excluding the papers related to medical experiments from the main list. There are a lot of it. Most of the false positives I marked was medical experiment paper.
3 Likes
Yeah, I concur, the RPM / 3D clinostat needs recognition system needs further training. But I think that’s the plan
1 Like
Thanks for your help! I updated the search results in this moment, but I’m still working with your information.
1 Like
I updated the search results, can you check if the order is better (I need your feedback before change some things)?.
1 Like