Add drawio.svg

Author: zhenghaoz

src/.vuepress/sidebar/en.ts

@@ -30,8 +30,8 @@ export const enSidebar = sidebar({
       "collapsible": true,
       "prefix": "concepts/",
       "children": [
-        "pipeline",
         "data-source",
+        "pipeline",
         {
           "text": "Recommenders",
           "icon": "recommend-fill",
@@ -45,8 +45,8 @@ export const enSidebar = sidebar({
             "external"
           ]
         },
-        "replacement",
         "ranking",
+        "replacement",
         "evaluation"
       ]
     },

src/docs/concepts/data-source.md

@@ -61,23 +61,4 @@ $$
 
 where $\mathbb{I}(i=\hat I^{(10)}_{u,i})$ is the $i$-th item in the top 10 recommendations.
 
-### Clustering Index and HNSW Index
-
-The clustering index is used to speed up searching user (item) neighbors, while the HNSW index is used to accelerate recommendations from matrix factorization. The quality of an index is evaluated by the recall:
-
-$$
-\text{recall@n}=\frac{|\text{top n by index}|}{|\text{top n by brute force}|}
-$$
-
-::: tip
-
-The recalls of indices are listed in the "System Status" section of the dashboard homepage.
-
-![](../../img/evaluation-neighbor-index-recall.jpeg =400x)
-![](../../img/evaluation-mf-index-recall.jpeg =400x)
-
-If the recall of an index is extremely low, consider turning off the index.
-
-:::
-
 [^1]: He, Xiangnan, et al. "Neural collaborative filtering." Proceedings of the 26th international conference on world wide web. 2017.

src/docs/concepts/evaluation.md

@@ -3,6 +3,12 @@ icon: pipeline
 ---
 # Pipeline
 
+![](../../img/pipeline.drawio.svg)
+
+## Default Pipeline
+
+![](../../img/default.drawio.svg)
+
 The workflow of Gorse is depicted in the following flowchart:
 
 ```mermaid
@@ -267,20 +273,3 @@ clock_error = "5s"
 # Server-side cache expire time. The default value is 10s.
 cache_expire = "10s"
 ```
-
-### Replacement
-
-In some cases, read items should be recommended to users again but with lower priorities. If `enable_replacement` is set, read items would be placed back to recommendations. The priority decay factors for positive feedbacks and read feedbacks are controlled by `positive_replacement_decay` and `read_replacement_decay`.
-
-```toml
-[recommend.replacement]
-
-# Replace historical items back to recommendations.
-enable_replacement = false
-
-# Decay the weights of replaced items from positive feedbacks.
-positive_replacement_decay = 0.8
-
-# Decay the weights of replaced items from read feedbacks.
-read_replacement_decay = 0.6
-```

src/docs/concepts/pipeline.md

@@ -47,23 +47,3 @@ x_i\sum^n_{j=1}v_{j,f}x_j-v_{i,f}x^2_i,&\text{if }\theta\text{ is }v_{i,f}
 $$
 
 Hyper-parameters are optimized by random search and the configuration `recommend.collaborative` is reused.
-
-[^1]: Rendle, Steffen, et al. "BPR: Bayesian personalized ranking from implicit feedback." Proceedings of the Twenty-Fifth Conference on Uncertainty in Artificial Intelligence. 2009.
-
-[^2]: He, Xiangnan, et al. "Fast matrix factorization for online recommendation with implicit feedback." Proceedings of the 39th International ACM SIGIR conference on Research and Development in Information Retrieval. 2016.
-
-[^3]: Rendle, Steffen. "Factorization machines." 2010 IEEE International conference on data mining. IEEE, 2010.
-
-[^4]: Zhang, Shuai, et al. "Deep learning based recommender system: A survey and new perspectives." ACM Computing Surveys (CSUR) 52.1 (2019): 1-38.
-
-[^5]: Bergstra, James, and Yoshua Bengio. "Random search for hyper-parameter optimization." Journal of machine learning research 13.2 (2012).
-
-[^6]: Auvolat, Alex, et al. "Clustering is efficient for approximate maximum inner product search." arXiv preprint arXiv:1507.05910 (2015).
-
-[^7]: Malkov, Yu A., and Dmitry A. Yashunin. "Efficient and robust approximate nearest neighbor search using hierarchical navigable small world graphs." IEEE transactions on pattern analysis and machine intelligence 42.4 (2018): 824-836.
-
-[^8]: Hu, Yifan, Yehuda Koren, and Chris Volinsky. "Collaborative filtering for implicit feedback datasets." 2008 Eighth IEEE international conference on data mining. Ieee, 2008.
-
-[^9]: Auvolat, Alex, et al. "Clustering is efficient for approximate maximum inner product search." arXiv preprint arXiv:1507.05910 (2015).
-
-[^10]: Malkov, Yu A., and Dmitry A. Yashunin. "Efficient and robust approximate nearest neighbor search using hierarchical navigable small world graphs." IEEE transactions on pattern analysis and machine intelligence 42.4 (2018): 824-836.

src/docs/concepts/ranking.md

@@ -2,3 +2,19 @@
 icon: download
 ---
 # Replacement
+
+
+In some cases, read items should be recommended to users again but with lower priorities. If `enable_replacement` is set, read items would be placed back to recommendations. The priority decay factors for positive feedbacks and read feedbacks are controlled by `positive_replacement_decay` and `read_replacement_decay`.
+
+```toml
+[recommend.replacement]
+
+# Replace historical items back to recommendations.
+enable_replacement = false
+
+# Decay the weights of replaced items from positive feedbacks.
+positive_replacement_decay = 0.8
+
+# Decay the weights of replaced items from read feedbacks.
+read_replacement_decay = 0.6
+```