Hello
\n\ni am reading this text\nhttp://www.manaalfaruqui.com/papers/naacl15-retrofitting.pdf
\n\nparagraf 2\n Retrofitting with Semantic Lexicons
\n\n In this case, we first train the word vectors independent of the information in the semantic lexicons and then retrofit them. Ψ is\n convex in Q and its solution can be found by solving a system of\n linear equations. To do so, we use an efficient iterative updating\n method (Bengio et al., 2006; Subramanya et al., 2010; Das and Petrov,\n 2011; Das and Smith, 2011). The vectors in Q are initialized to be\n equal to the vectors in Qˆ. We take the first derivative of Ψ with\n respect to one qi vector, and by equating it to zero arrive at the\n following online update: qi = P j:(i,j)∈E βijqj + αiqˆi P j:(i,j)∈E\n βij + αi (1) In practice, running this procedure for 10 iterations\n converges to changes in Euclidean distance of adjacent vertices of\n less than 10−2 . The retrofitting approach described above is modular;\n it can be applied to word vector representations obtained from any\n model as the updates in Eq. 1 are agnostic to the original vector\n training model objective.\nso my question:
\n\nanyone has somewhere a script doing this ?
\n\nthanks
", "messages": [ { "id": "425445", "postDate": "11/21/2018 16:07:08", "content": "Hello
\n\ni am reading this text\nhttp://www.manaalfaruqui.com/papers/naacl15-retrofitting.pdf
\n\nparagraf 2\n Retrofitting with Semantic Lexicons
\n\n In this case, we first train the word vectors independent of the information in the semantic lexicons and then retrofit them. Ψ is\n convex in Q and its solution can be found by solving a system of\n linear equations. To do so, we use an efficient iterative updating\n method (Bengio et al., 2006; Subramanya et al., 2010; Das and Petrov,\n 2011; Das and Smith, 2011). The vectors in Q are initialized to be\n equal to the vectors in Qˆ. We take the first derivative of Ψ with\n respect to one qi vector, and by equating it to zero arrive at the\n following online update: qi = P j:(i,j)∈E βijqj + αiqˆi P j:(i,j)∈E\n βij + αi (1) In practice, running this procedure for 10 iterations\n converges to changes in Euclidean distance of adjacent vertices of\n less than 10−2 . The retrofitting approach described above is modular;\n it can be applied to word vector representations obtained from any\n model as the updates in Eq. 1 are agnostic to the original vector\n training model objective.\nso my question:
\n\nanyone has somewhere a script doing this ?
\n\nthanks
", "rawMarkdown": "Hello\n\ni am reading this text\nhttp://www.manaalfaruqui.com/papers/naacl15-retrofitting.pdf\n\nparagraf 2\n Retrofitting with Semantic Lexicons\n\n In this case, we first train the word vectors independent of the information in the semantic lexicons and then retrofit them. Ψ is\n convex in Q and its solution can be found by solving a system of\n linear equations. To do so, we use an efficient iterative updating\n method (Bengio et al., 2006; Subramanya et al., 2010; Das and Petrov,\n 2011; Das and Smith, 2011). The vectors in Q are initialized to be\n equal to the vectors in Qˆ. We take the first derivative of Ψ with\n respect to one qi vector, and by equating it to zero arrive at the\n following online update: qi = P j:(i,j)∈E βijqj + αiqˆi P j:(i,j)∈E\n βij + αi (1) In practice, running this procedure for 10 iterations\n converges to changes in Euclidean distance of adjacent vertices of\n less than 10−2 . The retrofitting approach described above is modular;\n it can be applied to word vector representations obtained from any\n model as the updates in Eq. 1 are agnostic to the original vector\n training model objective.\n\nso my question:\n\nanyone has somewhere a script doing this ?\n\nthanks", "votes": null }, { "id": "425617", "postDate": "11/21/2018 21:13:30", "content": "Interesting paper, but not usable here anyway, since the semantic lexicons you'd need would be an external data source
", "rawMarkdown": "Interesting paper, but not usable here anyway, since the semantic lexicons you'd need would be an external data source", "votes": null } ], "comments": [ { "id": 425617, "author_name": "stevedraper", "author_url": "", "post_date": "11/21/2018 21:13:30", "content": "Interesting paper, but not usable here anyway, since the semantic lexicons you'd need would be an external data source
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "425445": "Hello\n\ni am reading this text\nhttp://www.manaalfaruqui.com/papers/naacl15-retrofitting.pdf\n\nparagraf 2\n Retrofitting with Semantic Lexicons\n\n In this case, we first train the word vectors independent of the information in the semantic lexicons and then retrofit them. Ψ is\n convex in Q and its solution can be found by solving a system of\n linear equations. To do so, we use an efficient iterative updating\n method (Bengio et al., 2006; Subramanya et al., 2010; Das and Petrov,\n 2011; Das and Smith, 2011). The vectors in Q are initialized to be\n equal to the vectors in Qˆ. We take the first derivative of Ψ with\n respect to one qi vector, and by equating it to zero arrive at the\n following online update: qi = P j:(i,j)∈E βijqj + αiqˆi P j:(i,j)∈E\n βij + αi (1) In practice, running this procedure for 10 iterations\n converges to changes in Euclidean distance of adjacent vertices of\n less than 10−2 . The retrofitting approach described above is modular;\n it can be applied to word vector representations obtained from any\n model as the updates in Eq. 1 are agnostic to the original vector\n training model objective.\n\nso my question:\n\nanyone has somewhere a script doing this ?\n\nthanks", "425617": "Interesting paper, but not usable here anyway, since the semantic lexicons you'd need would be an external data source" }, "source": "meta" }