ddiruscio/logseq
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
master
logseq/pages/hls__TSE-2023-11-0555_Proof_hi_1700083684930_0.md
T
ddiruscio 97fba407ce Major2
2025-06-05 22:07:12 +02:00

23 KiB
Raw Permalink Blame History

file:: TSE-2023-11-0555_Proof_hi_1700083684930_0.pdf file-path:: ../assets/TSE-2023-11-0555_Proof_hi_1700083684930_0.pdf

  • Links between issues and pull requests (PRs) assist GitHub developers in tackling technical challenges, gaining development inspiration, and improving repository maintenance. ls-type:: annotation hl-page:: 2 hl-color:: green id:: 658c4776-38d7-45cc-ad69-69d180881c00
  • these links are still insufficiently established ls-type:: annotation hl-page:: 2 hl-color:: purple id:: 658c477f-e523-4977-8b5f-24e790609c59
  • text similarity with additional information like issue size to predict issue-PR links, yet their effectiveness is unsatisfactory ls-type:: annotation hl-page:: 2 hl-color:: green id:: 658c48a7-5f90-455d-8a10-32db47b9ea01
  • PR Link Prediction ls-type:: annotation hl-page:: 2 hl-color:: green id:: 658c48b2-a0ea-47c6-874e-6f3fdd6e5f86
  • Auto IP Linker (AIPL) ls-type:: annotation hl-page:: 2 hl-color:: purple id:: 658c48cd-c468-4bc0-a520-347a0d9f15eb hl-stamp:: 1703692495878
  • GitHub sources with their relationships. ls-type:: annotation hl-page:: 2 hl-color:: green id:: 658c48e0-a6ed-4531-bd20-090a65f7f141
  • 4 types of GitHub sources related to issues and PRs(repositories, users, issues, PRs) as well as their relationships, and model them into task-specified heterogeneous graphs. ls-type:: annotation hl-page:: 2 hl-color:: blue id:: 658c4932-7add-4f68-9504-3bcf64f6f396
  • metapaths ls-type:: annotation hl-page:: 2 hl-color:: green id:: 658c493d-91ee-460d-a99e-f366109a6a64
  • metapath-based technology ls-type:: annotation hl-page:: 2 hl-color:: blue id:: 658c4943-4bb7-45e2-8d9e-85e176c3c8f6
  • submit issues and pull requests (PRs ls-type:: annotation hl-page:: 2 hl-color:: green id:: 65953950-2111-43ba-8039-b201a74971fe hl-stamp:: 1704278354819
  • More specifically, issues are used for reporting unexpected software behaviors, such as bugs and feature requests [5], [6], and repository contributors can submit PRs containing code changes to solve the challenges of issue report ls-type:: annotation hl-page:: 2 hl-color:: blue id:: 65953962-9df7-4c3a-b745-4554b3671bf6 hl-stamp:: 1704278383199
  • links between relevant PRs and issues to provide solutions, gain inspiration, prompt tracking, and expedite software developme ls-type:: annotation hl-page:: 2 hl-color:: blue id:: 6595396b-a84b-449a-93c8-6349dc9bfefc hl-stamp:: 1704278381545
  • establishing a link between issues and pull requests is not only necessary but also immensely beneficial for repository maintenance and evolution ls-type:: annotation hl-page:: 2 hl-color:: blue id:: 65953986-8608-4a88-b39b-521edb28febe
  • finding relevant solutions from the vast number of pull requests in a multitude of repositories, especially for large-scale repositories, is a labor-intensive and tedious tas ls-type:: annotation hl-page:: 2 hl-color:: blue id:: 659539b4-23de-47db-8d3e-a59daa7d7ce9
  • over half (54%) of PRs are not linked to an issue when submitted, despite the fact that contribution guidelines of their repositories recommend linking. ls-type:: annotation hl-page:: 2 hl-color:: blue id:: 659539df-f755-49d5-ae55-a43ad38e32e0 hl-stamp:: 1704279471236
  • To address the problem of lacking links between issues and PRs, several initial efforts based on open-source software have been dedicated to recommending links. ls-type:: annotation hl-page:: 2 hl-color:: green id:: 65953d9d-c9c3-4748-86ce-a732f1dd83ed hl-stamp:: 1704279466673
  • rely on the textual similarity between issues and PRs for link prediction ls-type:: annotation hl-page:: 2 hl-color:: purple id:: 65953da5-aa35-4a46-a140-4b62e9dbffec hl-stamp:: 1704279464427
  • issues and PRs with similar content may not necessarily be related, while those with dissimilar content could indeed be linked. ls-type:: annotation hl-page:: 2 hl-color:: blue id:: 65953dca-aeda-4c50-97e3-dbdaaa38819c
  • solely relying on textual similarity to recommend issue-PR links may be insufficient and less effective. ls-type:: annotation hl-page:: 2 hl-color:: green id:: 65953df2-b173-4c48-adc0-7142391d9b99 hl-stamp:: 1704279539633
  • textual similarity ls-type:: annotation hl-page:: 2 hl-color:: blue id:: 65953e12-6e6c-4612-bbd4-999842965395 hl-stamp:: 1704279573594
  • participant overlap ls-type:: annotation hl-page:: 2 hl-color:: blue id:: 65953e19-c6d5-4f07-b711-6c9259fa6f67
  • size of the issue report ls-type:: annotation hl-page:: 2 hl-color:: blue id:: 65953e1f-78e0-4bf4-93d1-8383258b3781
  • it is necessary to explore how to organize such useful and complex knowledge for the task of predicting issue-PR links on GitHub ls-type:: annotation hl-page:: 3 hl-color:: purple id:: 65953e40-09f6-43ad-8265-dc7d68d03de3 hl-stamp:: 1704279618923
  • eterogeneous graphs to organize their information. ls-type:: annotation hl-page:: 3 hl-color:: green id:: 6595721d-648c-432e-aa97-1304e52eb096
  • multi-types of nodes and edge ls-type:: annotation hl-page:: 3 hl-color:: green id:: 65957227-2be8-4298-947f-eae1887aa367
  • Fig. 3: The process of constructing a heterogeneous graph ls-type:: annotation hl-page:: 7 hl-color:: green id:: 65957238-48a3-48e3-9b9f-348d2d54c427
  • Figure 3, ls-type:: annotation hl-page:: 3 hl-color:: yellow id:: 6595730f-c19a-4453-a41f-077491bba291
  • [:span] ls-type:: annotation hl-page:: 7 hl-color:: green id:: 6595733c-36f4-444d-8fb7-f911dcc917fc hl-type:: area hl-stamp:: 1704293178574
  • the investigation of leveraging graph neural networks on heterogeneous graphs for issue-PR link prediction is rather limited ls-type:: annotation hl-page:: 3 hl-color:: purple id:: 65957a4d-42df-4906-bc53-097d9535c02f
  • ask-specified heterogeneous graph can represent all the aforementioned source information while also capturing their correlations. ls-type:: annotation hl-page:: 3 hl-color:: green id:: 65957a9d-5a38-445e-8e81-a4724fcdaf1d
  • metapath is a sequence of node and edge types within the specific schema, which can be regarded as paths of specific information transmission on heterogeneous graphs ls-type:: annotation hl-page:: 3 hl-color:: green id:: 65957aeb-6338-4a4b-a958-d083031dd810
  • metapath-based techniques can define the scope of crucial neighbors and extract information from them to avoid the information redundancy. ls-type:: annotation hl-page:: 3 hl-color:: green id:: 65957afe-c1d1-4a7c-9264-01c5551a4279
  • , a link classifier can predict the link between them ls-type:: annotation hl-page:: 3 hl-color:: blue id:: 65957b1f-e3f4-448a-bb59-4c18da1f5f63
  • given a pair of an issue and a PR, AIPL can suggest whether there can be a link ls-type:: annotation hl-page:: 3 hl-color:: blue id:: 65957b29-1722-498c-892c-e95253f62242
  • average performance of AIPL outperforms the other baseline, consisting of the iLinker [17], the A-M [15], random walk algorithm [27], R-GCN [28] and HAN [22], achieving average improvements of 24.97%, 23.99%,30.31%, and 27.98% in terms of accuracy, precision, recall, and F1-score. ls-type:: annotation hl-page:: 3 hl-color:: green id:: 65957b8a-bb56-4d09-a3a2-e75a679d434c
  • this is the first work to introduce heterogeneous graphs to anticipate GitHub issue-PR links ls-type:: annotation hl-page:: 3 hl-color:: purple id:: 659594b4-ee0f-4c4a-b63c-e41c58680f9c
  • we formulate 12 types of metapaths, enabling our approach can effectively learn embeddings of GitHub issues and PRs on a complex graph structure ls-type:: annotation hl-page:: 3 hl-color:: purple id:: 659594d1-edcb-40ea-850b-52db01f83473
  • preliminary relationships about issue-PR link prediction ls-type:: annotation hl-page:: 3 hl-color:: green id:: 659595ac-4154-408f-b7b6-b9e12f3b7a1f
  • metapath ls-type:: annotation hl-page:: 3 hl-color:: yellow id:: 6595966e-a8d4-4c79-9141-583b3633fa52
  • 29299 ls-type:: annotation hl-page:: 3 hl-color:: yellow id:: 65959707-79f7-4b35-bdc1-3fd8326416e3 hl-stamp:: 1704302487658
  • GitHub sources denotation ls-type:: annotation hl-page:: 3 hl-color:: blue id:: 6595973c-0c2a-4343-a0a0-01c81e94dbd1 hl-stamp:: 1704302399987
  • GitHub preliminary relationships denotation ls-type:: annotation hl-page:: 4 hl-color:: blue id:: 65959746-f4b1-4c48-887d-628ba63688f9
  • corresponding connections will be established if a user submits an issue or a pull request to a repository, or if the user contributes to or forks the repository. ls-type:: annotation hl-page:: 4 hl-color:: green id:: 6595975f-5ff5-4b79-9bfb-74bc1c5e53df
  • linked issues or linked PRs tend to share similar titles ls-type:: annotation hl-page:: 4 hl-color:: blue id:: 6595976e-a447-407a-953e-7b433f9c3043
  • PR #22248 with close titles ls-type:: annotation hl-page:: 4 hl-color:: yellow id:: 65959792-0254-4723-9952-99110bf3b1ac
  • e can also find repositories in the same subarea that share similar descriptions and tags ls-type:: annotation hl-page:: 4 hl-color:: green id:: 65959801-2d3e-46f3-94f9-edc9d346d02a
  • there is no direct edge linking the issues and PRs, since it is the link to be predicted. T ls-type:: annotation hl-page:: 4 hl-color:: blue id:: 65959821-c30e-4772-883c-ff6f55fe13b7
  • defined by submitting histories on GitHub and text similarity ls-type:: annotation hl-page:: 4 hl-color:: green id:: 65959848-5cbd-40d8-848e-495160e07c1e
  • Given a G where part of issue-PR node pairs(vI L, vP R L ) are provided with labels of linkage, ls-type:: annotation hl-page:: 4 hl-color:: green id:: 6595987d-6908-432a-8a33-7c01c8a6f980
  • redicting the existence of a link between issue node vI n and pull request node vP R n in the test set (vI T , vP R T ), i.e, f (vI n, vP R n ) → y (1) where f is the function to learn and y ∈ {0, 1} is the label of the issue-PR link. y = 1 indicates the existence of a link between vI n and vP R n , and y = 0, otherwise ls-type:: annotation hl-page:: 4 hl-color:: purple id:: 65959898-39d0-49da-a753-3b02620ea660 hl-stamp:: 1704302747031
  • we observe which source information can complement them and summarize the corresponding path information ls-type:: annotation hl-page:: 4 hl-color:: green id:: 65959998-9dfc-4683-b1dd-0bbab7eba05c
  • I-I. I ls-type:: annotation hl-page:: 4 hl-color:: yellow id:: 659599a5-02b9-47db-a81d-fd4f2fb94f92
  • similar issues tend to link to a PR, suggesting a need to exchange knowledge ls-type:: annotation hl-page:: 4 hl-color:: green id:: 659599c1-d0ce-4006-ba2b-859e4774059c
  • [:span] ls-type:: annotation hl-page:: 5 hl-color:: green id:: 659599fb-8ff2-46e9-978b-829bb68d6dac hl-type:: area hl-stamp:: 1704303098047
  • ssues with similar titles can complement the more detailed information, hence, we integrate the information from similar issues into target issue embeddings ls-type:: annotation hl-page:: 4 hl-color:: blue id:: 65959a51-4654-4cee-9fff-553b7978d13f
  • I-R-R-I to connect the issues of similar repositories to supplement more details of repository functions and tool scenarios ls-type:: annotation hl-page:: 4 hl-color:: green id:: 65968326-4ab4-4340-b47d-51d56f879b69
  • ata transmission in the repository facebook/reac ls-type:: annotation hl-page:: 4 hl-color:: green id:: 65968362-c4fe-4699-a0d8-b287a94382a4
  • I-U-I path to provide task-specific background information as well as contextual information for the corresponding issue. ls-type:: annotation hl-page:: 4 hl-color:: green id:: 6596838d-c419-42fd-9168-1566f8d6cb3b
  • submitters with similar development experienc ls-type:: annotation hl-page:: 4 hl-color:: green id:: 65968440-924f-4c4d-9e39-8eff172722e2
  • bugs in updating items on mobile chrome in different repositories ls-type:: annotation hl-page:: 4 hl-color:: green id:: 65968487-b2fe-475e-bf09-0661e7791407
  • users with similar development experiences may encounter similar questions across different repositories ls-type:: annotation hl-page:: 4 hl-color:: blue id:: 65968494-d3ee-415c-b69a-de54c3d9bc49
  • incorporate the issue with information from different users’ perspectives ls-type:: annotation hl-page:: 4 hl-color:: green id:: 659684a4-99e7-40f1-ad59-650940805334
  • issues from different repositories are connected through users who have contributed to both repositories. ls-type:: annotation hl-page:: 4 hl-color:: green id:: 65968543-b6b5-4565-9b40-defdefb10dfa
  • Leveraging the contributions of users who engage across various domains, we can bridge the gap and identify cross-domain repositories, facilitating the comprehensive integration of issue knowledge from cross-domain perspectives. ls-type:: annotation hl-page:: 5 hl-color:: green id:: 6596857d-bbd5-4922-95bf-0ed8fd2abdce
  • [:span] ls-type:: annotation hl-page:: 6 hl-color:: green id:: 659685b7-1ce2-4553-9d4c-753e290cd8c0 hl-type:: area hl-stamp:: 1704363446326
  • AIPL constructs a taskspecified heterogeneous graph by generating initial embeddings for all source nodes including issues, PRs, repositories, and users, and modeling relationships between nodes as edges. ls-type:: annotation hl-page:: 5 hl-color:: purple id:: 65968613-a566-4d24-9b99-3af0f6372bc9
  • aggregating information based on our heterogeneous graph and designed metapaths ls-type:: annotation hl-page:: 5 hl-color:: green id:: 6596862c-f372-4f5a-a4da-de1203adc2dc
  • ntra-metapath Aggregation ls-type:: annotation hl-page:: 5 hl-color:: blue id:: 65968644-032d-447a-abc7-0821ca3ae9e5
  • nter-metapath Aggregation ls-type:: annotation hl-page:: 5 hl-color:: blue id:: 6596864a-f125-4884-aa69-68205eb18fbc
  • 4 types of GitHub sources ls-type:: annotation hl-page:: 5 hl-color:: green id:: 6596867c-daeb-4870-b08a-b3792f750685
  • 8 types of relationships related to our task ls-type:: annotation hl-page:: 5 hl-color:: green id:: 65968681-f1ff-43f0-8bbc-7195fb2039bc
  • titles of issues and PRs ls-type:: annotation hl-page:: 5 hl-color:: green id:: 659686da-ecd3-4409-a86a-77e6270ba7f8
  • exclude the bodies and comments from consideration due to redundancy information ls-type:: annotation hl-page:: 5 hl-color:: blue id:: 659686e3-c249-4f75-abc5-3dba8d9edba3
  • Glove [32] pre-trained on 6 billion words of Wikipedia and Gigword corpus, to encode each title as a feature vector. ls-type:: annotation hl-page:: 5 hl-color:: green id:: 659686f0-1b37-4f46-9f75-1945f5bde315
  • 50-dimensional ls-type:: annotation hl-page:: 5 hl-color:: yellow id:: 65968700-6f86-4540-bb7d-55c401547f4d
  • Embeddings of Issue and PR ls-type:: annotation hl-page:: 5 hl-color:: purple id:: 6596872f-cb17-4ddd-b5d6-2d8f78dfc152
  • Embeddings of Repostitory ls-type:: annotation hl-page:: 5 hl-color:: purple id:: 65968735-27a0-4180-bd02-105f877590bb
  • Repository node embedding ls-type:: annotation hl-page:: 5 hl-color:: green id:: 6596897a-0038-4931-ac8e-3bc4158b2b24
  • Embeddings of Users. ls-type:: annotation hl-page:: 5 hl-color:: purple id:: 659689e2-9e05-4dad-9b1a-8ced8d88348f
  • Each element within this vector is populated with a randomly selected integer, drawn from a range between 1 and 10 ls-type:: annotation hl-page:: 5 hl-color:: yellow id:: 65968a31-ad3a-436f-b8d0-c18f2e28fee3
  • task-specific heterogeneous graph ls-type:: annotation hl-page:: 6 hl-color:: yellow id:: 65968a5b-eaa6-4479-b476-7112e4224831
  • relationships between similar repositories, issues, and PRs are necessary to construct ls-type:: annotation hl-page:: 6 hl-color:: green id:: 65968b39-19bb-42c9-9591-14aa70d11177
  • cosine similarity between node embeddings of VR learned by the method of Section 3.1.1. ls-type:: annotation hl-page:: 6 hl-color:: yellow id:: 65968b48-6612-4b6b-9278-59d1c314675d
  • Having constructed all node embedding and edges, we generate a heterogeneous graph encompassing various sources and relationships on GitHub. ls-type:: annotation hl-page:: 6 hl-color:: green id:: 65968ba8-e7fe-4491-af4e-03e09c98b2e5
  • For example, the embedding dimension of VU is 10, while the embedding dimensions of other node types are defined as 50. ls-type:: annotation hl-page:: 6 hl-color:: yellow id:: 65968cc7-956f-485d-93d3-660ab330b1cd hl-stamp:: 1704365257360
  • After applying this operation, all nodes’ projected embeddings share the same dimension defined as 10 referenced [35], which improves the aggregation process of the next stage ls-type:: annotation hl-page:: 6 hl-color:: yellow id:: 65968d86-3e1d-48d5-8c5b-a09607a81ab8
  • Z. Wu, S. Pan, F. Chen, G. Long, C. Zhang, and P. S. Yu, “A comprehensive survey on graph neural networks,” IEEE Transactions on Neural Networks and Learning Systems, vol. 32, pp. 4–24, 2019 ls-type:: annotation hl-page:: 15 hl-color:: green id:: 65968dac-b39a-415e-a337-0685443871f7
  • earn the comprehensive embeddings of issues and PRs ls-type:: annotation hl-page:: 6 hl-color:: blue id:: 65968eaa-1a2e-4b47-925e-0269319b0bdd hl-stamp:: 1704365742235
  • [:span] ls-type:: annotation hl-page:: 7 hl-color:: green id:: 65968edb-6147-4c5f-96ec-ffb4d0825cd5 hl-type:: area hl-stamp:: 1704365785085
  • o integrate the information under each metapath ls-type:: annotation hl-page:: 7 hl-color:: yellow id:: 65968f61-185f-4a41-a3ac-f0e1757ef85a
  • nalysis 12 ls-type:: annotation hl-page:: 7 hl-color:: red id:: 65968f85-77f2-47e4-9383-329524922a9e
  • Metapath Aggregation ls-type:: annotation hl-page:: 6 hl-color:: yellow id:: 65968ff9-fcb8-4a76-b187-2a47a5dc668a
  • the set of nodes {repository concord-consortium /codap, user Maaartinus, repository folio-org/stripes-core} is the metapath-based intermediate nodes of node PR #103 ls-type:: annotation hl-page:: 6 hl-color:: green id:: 6596cbb4-18f7-4375-84af-8224791c243b
  • redict the relationships between issues and PRs ls-type:: annotation hl-page:: 8 hl-color:: green id:: 6596ccb4-bd4d-40b8-a6d7-fe91430c7a92
  • sigmoid ls-type:: annotation hl-page:: 8 hl-color:: green id:: 6596cccb-dac1-4cd1-b8b8-556db3b86736
  • overall forward propagation process is shown in Algorithm 1. ls-type:: annotation hl-page:: 8 hl-color:: green id:: 6596cce8-c593-4a54-b572-d7e4074cb88a
  • sk-specified heterogeneous graph G, metapaths P and the number of attention heads K. ls-type:: annotation hl-page:: 8 hl-color:: yellow id:: 6596ccf0-a01e-4520-a27f-f3fbae40da0e hl-stamp:: 1704381683349
  • RQ1 Can our proposed AIPL effectively predict issue-PR links on GitHub? ls-type:: annotation hl-page:: 9 hl-color:: blue id:: 6596cd2f-443d-44be-b1e4-ed3e6acf65e6 hl-stamp:: 1704381747340
  • RQ2 Can AIPL be effective when specific portions of the heterogeneous graph or metapaths are removed? ls-type:: annotation hl-page:: 9 hl-color:: yellow id:: 6596cd40-4729-46b3-bdb7-6cb147aac7ab hl-stamp:: 1704381763243
  • RQ3 How does the performance of AIPL vary when different thresholds or numbers of attention heads are applied? ls-type:: annotation hl-page:: 9 hl-color:: blue id:: 6596cd99-5485-4b7c-8750-512545f2e07f
  • lthough some studies have released datasets for similar tasks, they have primarily focused on predicting links between issues and PRs within the same repositories. In contrast, our task aims to link issues and PRs across various repositories in GitHub. ls-type:: annotation hl-page:: 9 hl-color:: purple id:: 6596ce97-647e-436f-a177-0e76b1ea9383
  • multiple data sources, including information from issues, PRs, repositories, and users, which are absent in prior datasets. Therefore, we have chosen to construct new datasets, ls-type:: annotation hl-page:: 9 hl-color:: green id:: 6596ced6-71f7-4a56-a760-ce415d2c5123
  • facebook/react and vuejs/vue ls-type:: annotation hl-page:: 9 hl-color:: green id:: 6596ceff-ffc9-4d36-9e09-7cf22c9f792d hl-stamp:: 1704382209179
    1. These two repositories host the most stars on GitHub, apart from some recommended lists, tutorials, and books ls-type:: annotation hl-page:: 9 hl-color:: blue id:: 6596cf46-a37b-42a3-b85b-831a230fd6c8
    1. they contain plenty of issues and PRs for our research (facebook/react host 27,229 issues and PRs and vuejs/vue host 13,064 by August 2023.) ls-type:: annotation hl-page:: 9 hl-color:: green id:: 6596cf5a-9b46-4597-a6e7-1dad1cd2ed14
    1. active communities are still dedicated to maintaining and developing these repositories and our work may help their contributors and subscribers. ls-type:: annotation hl-page:: 9 hl-color:: green id:: 6596cf69-a5b8-4690-b6de-2b96339afae4
  • randomly select 19 repositories from the above candidates ls-type:: annotation hl-page:: 9 hl-color:: green id:: 6596d13d-6a1f-4feb-8e53-e80127617bd2
  • we filter out non-English issues and PRs ls-type:: annotation hl-page:: 9 hl-color:: green id:: 6596d37a-a15b-435c-b830-55b5cc888e68
  • we remove the issues and PRs that are submitted by bots, since these issues and PRs might lack essential content and primarily serve as automated notifications ls-type:: annotation hl-page:: 9 hl-color:: green id:: 6596d38d-aa87-42aa-aa64-55ef35bc03f9 hl-stamp:: 1704383375451
  • e delete duplicate issues and PRs to ensure the uniqueness of the items in this dataset ls-type:: annotation hl-page:: 9 hl-color:: green id:: 6596d398-7599-4ac0-9f83-bc4591b4801f
  • proceed ls-type:: annotation hl-page:: 9 hl-color:: red id:: 6596d3f2-03ed-4fdf-b523-c005696fd88a
  • we randomly match an issue from the aforementioned positive samples with a PR ls-type:: annotation hl-page:: 9 hl-color:: green id:: 6596d408-3b2e-4f4c-b1f1-af0c8e30bdaf
  • we introduce manual inspection to ensure the quality of our dataset ls-type:: annotation hl-page:: 9 hl-color:: blue id:: 6596d412-c1f0-4620-bb6d-9cdf0461a093
  • Cohen’s kappa agreement is calculated to be 0.7332, indicating an acceptable level of consistency in the results. ls-type:: annotation hl-page:: 9 hl-color:: green id:: 6596d445-91f8-4475-8373-ccd62ab81bd2
  • . Finally, all the authors double-check the annotation results to ensure a 100% agreement rate ls-type:: annotation hl-page:: 10 hl-color:: green id:: 6596d48e-17dd-4304-ac9b-7037a0dbd6c1

  • of {I-PR} 8,457 3,655

    ls-type:: annotation hl-page:: 9 hl-color:: green id:: 6596d510-a686-459e-85a2-244b6e467914
  • The remaining data and linkage labels are used to train our model ls-type:: annotation hl-page:: 10 hl-color:: green id:: 6596d53d-abd6-4392-86e1-26ee2bb72236
  • 90/10 chronological split on the remaining dataset for the training set and validation set ls-type:: annotation hl-page:: 10 hl-color:: green id:: 6596d547-44dd-41f2-bcfb-7e277a528094
  • hen, given a target node such as an issue, the algorithm would explore the graph by performing random walks to generate a sequence of PR nodes. ls-type:: annotation hl-page:: 10 hl-color:: green id:: 6596d66d-8948-4869-8f13-6f18db14fe30
  • embeddings ls-type:: annotation hl-page:: 10 hl-color:: green id:: 6596d688-6542-487b-9a75-7148e33a4547
  • In cases where the content of certain repositories, issues, or PRs is empty, we initialize their 50-dimensional vectors with random values [30], [38], [39]. ls-type:: annotation hl-page:: 10 hl-color:: yellow id:: 6596d7e3-b4d9-4358-bb82-5bd8ada7bc6c
  • we plan to take our approach one step further by employing a larger-scale dataset to predict links from small and new repositorie ls-type:: annotation hl-page:: 14 hl-color:: green id:: 6596d8ae-f372-4ca3-abfb-9db01755ea4d
Reference in New Issue View Git Blame Copy Permalink