|This article is of interest to the following WikiProjects:|
This is wrong, Anderson and myself are right. Markov chains
The formal description is faulty as usual. For example, the activation level is in [0,1], but the algorithm presented may activate nodes to more than 1.0 There are other problems. The only sound descriptions I know are Anderson 1983 and my own PhD diss. My diss. also contains a comparison with Markov chains. When I have time I'll try to improve this article. Others please feel free. My diss. can be found here is here: https://www.box.com/shared/static/m4f8icz86m3dkv7r9dpa.pdf Marius63 (talk) 11:39, 11 July 2013 (UTC)
Also these papers are relevant to SA and Semantic Web:
Relation to Markov chain?
The spreading activation algorithm, as described in this article, appears to be similar to a Markov chain, and, based on my current reading/misunderstanding of this article, appears to be almost identical, except for various odd tweaks (like the minimal F to fire). To be precise, if is the weight matrix from this article, and R is the decay factor from this article, then, from what I can tell,
seems to give a very nearly equivalent Markov chain. (Although the , as I wrote it here, is mis-normalized, which might be why the activation of the neurons needs to be clamped to zero, one, since this mis-normalization will propagate through. Proper normalization wouldn't require clamping). FWIW, the symbol is the Kronecker delta function.
To summarize: the only differences I can spot are:
- Use of the minimal firing factor F
- Clamping (which wouldn't be needed if the system was normalized)
Am I missing something else? I ask, because these two differences alone, although they alter the Markov chain result somewhat, really wouldn't make much of a qualitative difference. And if there's no qualtitative difference, it begs the question: why bother with this (what would the theoretical reasons be?) as opposed to the far-more-widely-known, better-understood Markov chain? linas (talk) 15:36, 30 September 2008 (UTC)
This article should have: a good lead section, clear structure, balanced coverage, neutral content, and reliable sources.
Elements that make this article good: • Content is neutral. • No grammatical or spelling problems. • There is no warning banner at the top of the article. • The first two sentences of the lead are good enough for the leading paragraph, cut out the last three sentences (or put them under another section). • The links throughout the article (though there aren’t very many) do work. • The notes section is decent. • All of the content is there, but there needs to be more under each section.
Elements that need improvement: • Some of the language in the first paragraph is complex. I have no idea what the word iteratively means. Clarify some words. • The two sentences in the lead that start with “spreading activation” should be placed somewhere else. • This article is somewhat short and does not provide much information. • More references needed (there are only two listed). • More reliable sources needed. • There should be more headings under the contents. Maybe add some history, developments, practical applications, and more examples. The more examples, the more understandable the article will be. • The Algorithm section is very wordy and confusing (maybe add a picture or some type of visual). The bracket letters are confusing. • There needs to be fewer sentences in the lead paragraph. • The structure needs some work. • More subheads, images, and diagrams needed. • The more external links the better. • You need to put the explanation of the example on the left side and put the picture on the right side of the page. You need to visually balance the page. • This article is not balanced very well, it only has three sections. The Algorithm section is the biggest, so space it out more, or add more information to other sections or add some more sections. • There should be more links throughout the article so people will be able to link more. The word “source nodes” should be highlighted. • I know that the example section is simply just a picture and that there is an explanation on the bottom of it, but you should still clarify what is happening in the model. • You need to simplify the steps under Algorithm, I know this might be somewhat difficult but you don’t want people to get lost in this article (which I did after looking at the steps). • The steps should be referenced. There should be no unsourced opinions in the article.
Peer Review: The lead section was a little confusing. I would try to put a little more explanation in this section, especially if the subject is something that might be hard to grasp. The structure of the article, as a whole, was fairly clear and understandable. Although, there were times where I had no idea what the article was getting at. It would be good to go back through and read the article as if you knew nothing about it before, and if you can't understand then it should be fixed. It was well balanced; however, more subsections should have been added rather than just having three main sections so that the reader fully understands the concept behind “Spreading Activation” when they are done reading. There is only a "Spreading Activation" section, "Algorithm" section, and "Examples" section. These did help in explaining the topic, but not completely. Adding more detail and examples under each of these sections would also help the reader. The coverage of the article is neutral; however, I think it would be good to have had more sources. The sources provided are reliable, but with more information, comes more sources. There is no warning banner at the top, which is good. As mentioned earlier, the language in the lead section is a bit confusing. It is the beginning of the article so it shouldn't be that way. Also, some of the sentences are too long. Too many thoughts are in one sentence, and can lead to confusion. One example of this is in the sentence: “The search process is initiated by labeling a set of source nodes (e.g. concepts in a semantic network) with weights or "activation" and then iteratively propagating or "spreading" that activation out to other nodes linked to the source nodes.” Parts of this sentence could be explained separately. The sentence maybe should have been broken up into pieces and explained more. There are no un-sourced opinions or value statements and no section is overly long compared to the others. These things definitely help with the quality of the article. Lie46840 (talk) 02:09, 17 October 2012 (UTC)
Peer review: The beginning part is a bit confusing to me. Explanation could go a bit further through out the entire article. Although the articles structure was clear I need more information. This subject is foreign to me completely and by the end of the article I still didn't know what is was exactly. This article simply needed more references. Detail could make this article everything. If it was expanding by a clearer beginning it would add information to a reader who had no cognition background. If the author added more examples to the example section that would also add clarity. There are terms that are never fully explained as well like neural networks. If that section was flushed out a little more it would clear and concise to the reader. AS well as the term semantic networks. Although you can click on the link and simply go to the article about the networks it would be easier for the reader to have it explain in regards to the “spreading activation” article. I also do not understand the algorithm section AT ALL. I’m terrible at math and need things explained very slowly and this article needs to make that a little easier to understand. All sources are correctly referenced and no information is stated without a source. Gretchen Gosser
Example figure issues
This is super pedantic, but I think one of the numbers in the figure example is wrong. I think the activation on node2 should be 0.765, i.e. 77% not 76%. Also, there is no mention of the firing threshold in this example. Cai (talk) 18:18, 6 February 2018 (UTC)
Algorithm section unclear
There is a lack of clarity with the algorithm section. For example, it does not state at what point the nodes "marked for firing on the next … cycle" will actually fire. It also does not specify whether a node [j] in step 3 must be unfired. I would update it myself, but there is no citation given for the algorithm. Cai (talk) 18:25, 6 February 2018 (UTC)