Neural Latent Semantic Indexing Algorithm
Jump to navigation
Jump to search
A Neural Latent Semantic Indexing Algorithm is an Information Retrieval Algorithm that incorporates Latent Semantic Indexing into a Neural Network Training Algorithm.
- Context:
- It can be implemented by a Neural Latent Semantic Indexing System to solve a Neural Latent Semantic Indexing Task.
- Example(s):
- Counter-Example(s):
- See: Neural Semantic Indexing Algorithm, Supervised Semantic Indexing (SSI) Algorithm Semantic Analysis, Vectorial Semantics, Artificial Neural Network, Synonymy, Polysemy.
References
2017
- (Quispe et al., 2017) ⇒ Oscar Quispe, Alexander Ocsa, and Ricardo Coronado. (2017). “Latent Semantic Indexing and Convolutional Neural Network for Multi-label and Multi-class Text Classification.” In: Proceedings of IEEE Latin American Conference on Computational Intelligence (LA-CCI 2017). doi:10.1109/LA-CCI.2017.8285711
2014
- (Shen et al., 2014) ⇒ Yelong Shen, Xiaodong He, Jianfeng Gao, Li Deng, and Grégoire Mesnil. (2014). “A Latent Semantic Model with Convolutional-Pooling Structure for Information Retrieval.” In: Proceedings of the 23rd ACM International Conference on Conference on Information and Knowledge Management. ISBN:978-1-4503-2598-1 doi:10.1145/2661829.2661935
- QUOTE: In this study, we develop a new latent semantic model based on the convolutional neural network with convolution-pooling structure, called the convolutional latent semantic model (CLSM), to capture the important contextual information for latent semantic modeling. Instead of using the input representation based on bag-of-words, the new model views a query or a document [1] as a sequence of words with rich contextual structure, and it retains maximal contextual information in its projected latent semantic representation. The CLSM first projects each word within its context to a low—dimensional continuous feature vector, which directly captures the contextual features at the word n-gram level (detailed in section 3.3). Second, instead of summing over all word n-gram features uniformly, the CLSM discovers and aggregates only the salient semantic concepts to form a sentence-level feature vector (detailed in section 3.4). Then, the sentence-level feature vector is further fed to a regular feed—forward neural network, which performs a non—linear transformation, to extract high—level semantic information of the word sequence. In training, the parameters of the CLSM is learned on clickthrough data.
- ↑ In modern search engines, a Web document is described by multiple fields [12][38], including title, body, anchor text, etc. In our experiments, we only used the title field of a Web document for ranking. In addition to providing simplicity for fast experimentation, our decision is motivated by the observation that the title field gives better single-field retrieval result than body, although it is much shotter (as shown in Table 4). Thus it can serve as a reasonable baseline in our experiments. Nevertheless, our methods are not limited to the title field, and can be easily applied to the multi—field description.
1996
- (Syu et al., 1996) ⇒ Inien Syu, S. D. Lang, and Narsingh Deo. (1996). “Incorporating Latent Semantic Indexing Into a Neural Network Model for Information Retrieval.” In: Proceedings of the fifth International Conference on Information and knowledge management. ISBN:0-89791-873-8 doi:10.1145/238355.238475